Seminars

DIS Seminar Series

This web page presents the DIS seminar series. The aim of the series is to present ongoing work and new research ideas in the area of databases and it targets researchers and students who are interested in databases.

Archive of Seminar Slides

For slides of the past seminars, please click here

Schedule

The current schedule of the seminars is as follows. Click on a speaker’s name to get more detailed information about his/her seminar.


Note Date Time Speaker Title Slides
Oct 3, 2013 11:00-12:00pm Matthias Renz Modeling and Querying Uncertain Spatio-Temporal Data
Sep 2, 2013 3:00-4:00pm Patty Kostkova From Online Evidence to Social Networking and Digital Epidemiology – new challenges and opportunities for ehealth
Jun 27, 2013 3:00-4:00pm Olawale Titiloye Optimization by quantum annealing for the graph coloring problem
Feb 13, 2013 11:00-12:00 Victor Codina Semantically-Enhanced Pre-Filtering for Context-Aware Recommender Systems
Feb 4, 2013 11:00-12:00 Fabio Persia Discovering the Top-k Unexplained Sequences in Time-Stamped Observation Data
Jan 30, 2013 11:00-12:00 Maciej Piernik Pattern-based XML mining
Jan 11, 2013 11:00-12:00 Shuaiqiang Wang Ranking-Based Collaborative Filtering
Dec 19, 2012 4pm-5:30pm Anton Dignös Temporal Alignment
Nov 26, 2012 11:00am-12:00pm Johann-Christoph Freytag Stratosphere – Above the Clouds or Using Map/Reduce for Implementing Next-Generation Database Systems
Nov 19, 2012 11:00am-12:00pm Jane Yau A mobile context-aware learning schedule framework with Java learning objects
Sep 26, 2012 11:00am-12:00pm Martin Theobald Efficient Query Processing in Uncertain RDF Knowledge Bases
Sep 18, 2012 11:00am-12:00pm Irena Koprinska Recommender Systems for Online Dating
Sep 5, 2012 11:00am-12:00pm Andrej Košir What can pattern recognition tells us about modelling recommender systems
Sep 4, 2012 15:00am-16:00pm Marko Tkalčič Affect in recommender systems
Aug 23, 2012 11:00am-12:00pm Mohamed‐Amine Baazizi Static Analysis for Optimizing the Update of Large Temporal XML Documents
July 26, 2012 11:00am-12:00pm Adolfo Urrutia Outlining recommendation zones for taxi drivers using self-organizing maps
July 10, 2012 11:00am-12:00pm Victor Codina Semantically-Enhanced, Context-Aware Recommendation Algorithms
July 4, 2012 11:30am-12:30pm Theodoros Chondrogiannis A Geography-aware Service Overlay Network for Managing Moving Objects
Feb 7, 2012 11am-12pm Boris Glavic Provenance for Databases and Data Integration

Jan 27, 2012 3pm-4pm Arturas Mazeika Visual Analytics of Evolving Named Entities

Dec 1, 2011 2pm-3pm Anton Dignös Curbing Time Intervals in Database Systems

Nov 21, 2011 11am-12pm Markus Schedl Inferring Music Similarity from Content and Context
July 20 11AM-12PM Alexandros Karatzoglou Collaborative Ranking for Recommender Systems
July 8 11AM-12PM Wolfgang Nejdl Web Science @ L3S – Interdisciplinary Research Challenges
July 1st 11AM-12PM Neal Lathia Mobile Recommendations for Smart Cities
June 21 2PM-3PM Christian Koncilia Advanced Data Warehouse Analysis
June 1st 11AM-12PM Claudio Eccher, Chiara Ghidini, and Enrico Maria Piras Modeling the information flow process in hospitals: an experience in an oncology ward
June 1st 10AM-11AM Floriana Grasso Computational Argumentation for Digital Interventions
May 26 3PM-4PM Periklis Andritsos What can clusters reveal about your data: sieving through unstructured information sources
May 25 3PM-4PM Benjamin Gufler Handling data skew in MapReduce
May 19 3PM-4PM Christian von der Weth FAST: Friends Augmented Search Techniques – System Design & Data-Management Issues
May 18 3PM-4PM Saad Malik Combining Granularity-Based Topic-Dependent and Topic-Independent Evidences for Opinion Detection
January 21 11AM-12PM Giovanni Semeraro Content-based Recommender Systems: problems, challenges and research
 directions
December 1 2:30PM-4:30PM Martin Theobald Interactive Reasoning in Large and Uncertain RDF Knowledge Bases

November 29 11AM-12PM Iván Cantador Bringing Semantics to Folksonomies: Application to Item Recommendation
September 23 3PM-4PM Shlomo Berkovsky Group-Based Recipe Recommendations: Analysis of Data Aggregation Strategies
July 1 2PM-3PM Neil Rubens Recommending Learning Actively
June 15-17 11AM-12PM Robin Burke Social Computing
May 4 9:30AM-10:30AM Florian Michahelles Towards a More Tangible Search of Information: Internet of Things
April 1 2PM-3PM Floriano Zini Evaluation of Scheduling and Replica Optimization Strategies for a Data Grid
March 18 2PM-3PM Juozas Gordevicius Ranking of Evolving Stories Through Meta-Aggregation
February 17 3PM-4PM Augsten Nikolaus TASM: Top-k Approximate Subtree Matching
December 11 11AM-12AM Bhaskar Mehta Recommendations at Google
December 9 1PM-3PM Sven Helmer XML Message Management
December 2 9:30AM-10:30AM Peer Kröger How to find clusters in high-dimensional spaces: problems, solutions, and perspectives
December 1 2PM-3PM Alejandro Vaisman Integrating Spatial, OLAP, and Moving Object Data
November 27 9AM-10AM Alexander Felfernig Effective Knowledge Engineering for Constraint-based Systems
November 16 11AM-12AM Sherif Sakr GraphREL: A relational approach for scalable processing of subgraph queries
November 10 9AM-10AM Markus Zanker Harnessing Geotagged Resources for Web Personalization
November 3 9AM-10AM Lior Rokach Active Learning for Preferences Elicitation in Recommender Systems
October 21 9AM-10AM Boris Glavic Integrating Data Provenance Support in Database Systems

June 26 11:00am-12:00pm Robin Burke Robust Recommender Systems
June 22 9:00am-10:00am Federica Paci An integrated Digital Identity and Access Management Solution for Business Processes
June 3 3:00pm-4:00pm Bjørn Zenker ROSE
March 20 3:00pm-4:00pm Gerasimos Marketos Analyzing Trajectory Data
February 17 11:00am-12:00pm Ciro Cattuto Measuring and Grounding Similarity from Social Annotations
February 16 11:00am-12:00pm Gedas Adomavicius Overcoming Accuracy-Diversity Tradeoff in Recommender Systems
January 26 11:00am-12:00pm Arta Dilo Data structures for progressive transfer of 2D spatial data in a client-server environment
January 23 3:00pm-4:00pm Arturas Mazeika Bachelor: Minimizing Stochastic Coherence of Web Archives
January 21 3:00pm-4:00pm Alejandro Vaisman Constraint Evaluation in Categorical Sequential Pattern Mining
December 9 2:00pm-3:00pm Cosimo Palmisano A Data Mining Strategy For Targeted Sales Actions: A Case Study

November 25 2:00pm-3:00pm Paolo Cremonesi A Large-Scale Recommender System for an IP Television Service Provider
November 3 11:00am-12:00pm Thomas Roth-Berghofer A Unifying View on Explanation-aware Computing – Supporting the Use of Complex Information Systems
October 29 11:30am-12:30pm Daniele Quercia Mobile Content-Sharing Applications
October 13 3:00pm-4:00pm Christian Müller Recommendations Based on Speech Classification (and examples of what recommender systems can learn from signal processing)
September 5 3:00pm-4:00pm Curtis Dyreson About Closeness in XML and Supporting Proscriptive Metadata in an XML DBMS

Previous

Speaker and Talk Details

Matthias Renz

TIME: Oct 3, 2013, 11:00-12:00

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Matthias Renz, LMU Munich, Germanyy

TALK TITLE: Modeling and Querying Uncertain Spatio-Temporal Data

TALK ABSTRACT: The advances in sensing and telecommunication technologies allow the collection and management of vast amounts of spatio-temporal data combining location and time information. Due to physical and resource limitations of data collection devices (e.g., RFID readers, GPS receivers and other sensors) data are typically collected only at discrete points of time. In-between these discrete time instances, the positions of tracked moving objects are uncertain. The problem of modeling and managing uncertain data has received a great deal of interest, due to its manifold applications in spatial, temporal, multimedia and sensor databases. There exists a wide range of work covering spatial uncertainty in the static (snapshot) case, where only one point of time is considered. In contrast, the problem of modeling and querying uncertain spatio-temporal data has only been treated as a simple extension of the spatial case, disregarding time dependencies between consecutive timestamps. In this talk, i will present a framework for efficiently modeling and querying uncertain spatio-temporal data. The key idea of this approach is to model possible object trajectories by stochastic processes. This approach has three major advantages over previous work. First it allows answering queries in accordance with the possible worlds model based on the paradigm of probabilistic query processing. Second, dependencies between object locations at consecutive points in time are taken into account. And third it is possible to reduce all queries on this model to simple matrix multiplications. Based on these concepts, in the first part of my talk i will introduce efficient solutions for different probabilistic spatio-temporal queries. In the second part, i will propose novel approximation techniques in order to probabilistically bound the uncertain movement of objects; these techniques allow for efficient and effective filtering during query evaluation using an hierarchical index structure.

REFERENCE PERSON: Johann Gamper


Patty Kostkova

TIME: Sep 2, 2013, 3:00-4:00pm

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Patty Kostkova, UCL, London, UK, and ISI Foundation, Turin, Italy

TALK TITLE: From Online Evidence to Social Networking and Digital Epidemiology – new challenges and opportunities for ehealth

TALK ABSTRACT: The massive explosion of medical evidence on the Internet and mobile devices and increased citizens’ involvement in management of their health changed the delivery of healthcare and enhanced the potential of international public health surveillance.

Drawing from a body of research conducted at the City ehealth Research Centre in 2002-12 and at UCL and ISI Foundation, this presentation will outline the issues surrounding the dissemination of medical evidence to public and professionals, implications for public health surveillance and early warning systems and user engagement. Understanding public and professionals’ information needs from Internet search weblogs provides evidence for informing policy and drives content improvements of medical sites and their priorities. Serious games for health became established as an alternative method for teaching all generations about their health and personal conditions. Research into evaluation methods and educational effectiveness of serious games is an interdisciplinary domain spanning across gaming, user modelling, pedagogy, information science and user engagement.  Finally, social networking with increasing amount of user-generated content from social media and participatory surveillance systems provides readily available source of real-time epidemic intelligence information for early warning systems and novel channels for risk communication during public health emergencies.

In this talk, results from researching unique longitudinal user search datasets from the National electronic library of Infection and National Resource of Infection Control (NeLI/NRIC) portals will be presented together with our experience from multi-language educational health initiative Edugames4All: www.edugames4all.org winning the Best Student Paper GALA Award at VS-Games 2012; a finalist project of the UK IT Industry Award 2012 – a collaborative wiki technology for training epidemiologist (FEMwiki www.femwiki.com);  and the EHI 2012 finalist project investigating the potential of Twitter as an outbreak detection tool and effective risk communication channel during swine flu 2009.

SPEAKER’S BIO: Dr. Patty Kostkova is currently the Principal Research Associate for ehealth at the Department of Computer Science, University College London (UCL) and holds a Research Scientist post at the ISI Foundation in Italy. Patty is also a Special Lecturer at the School or Arts and Social Sciences, City University, London delivering an interdisciplinary module on Social Media for postgraduate social science and informatics students.
Until 2012, she was Reader and the Head of City eHealth Research Centre (CeRC) at City University, London, UK (http://www.city.ac.uk/health/research/research-areas/city-ehealth-research-centre/staff/dr-patty-kostkova). With an MSc and PhD degrees in computer science and an extensive international experience at public health agencies such as WHO and ECDC, Patty built up CeRC into a thriving multidisciplinary research centre collaborating with international partners and funding bodies including ECDC, WHO, HPA, EC and the DH. CeRC piloted a novel model enabling direct technology transfer of a user-driven high impact research through a family of real-world online services for medical professionals including the National Resource for Infection Control (NRIC), ECDC training resource FEM Wiki and educational games for children edugames4all.
Patty’s research successes were recognized by a number of prestigious awards including the UK IT Industry Awards 2012 – finalist in Category „IT project demonstrating most effective use of collaborative technology, EHI (Ehealth Insider) 2012 Prize – finalist in Category “Best Use of social media in healthcare” and The Best Interdisciplinary Research Project in the 4th Annual City University Research Competition.
In the recent years, she was appointed a consultant at WHO, ECDC and Foundation Merieux. In addition to Patty’s Advisory Board memberships including ECDC Knowledge Management Working Group and the NHS National Knowledge Service TB Pilot project, she established and chaired an interdisciplinary international eHealth conference in 2009-11 and in May 2013 chaired the WWW 2013 Workshop on Public Health in the Digital Age (PHDA 2013).
Regular invited and keynote speaker at prestigious institutions and international conferences, in March 2011, she was invited by BMJ as the “Idea Champion” to present her latest work on the potential of social media for epidemic intelligence receiving a wide media coverage including a BMJ scientific film: Medical Innovations: Twitter Epidemics
Patty published over 90 peer-reviewed papers, a book chapter and edited a number of journals. Her research was covered by international media including the Medi1TV, BBC, AFP, the Vancouver Sun, the Malaysian Insider and other.

REFERENCE PERSON: Francesco Ricci (http://www.inf.unibz.it/~ricci/)

Olawale Titiloye, Manchester Metropolitan University

TIME: Thursday, 27 June 2013, 3:00pm

ABSTRACT: Due to the intractability of exact solvers for large sizes of several combinatorial optimization problems, metaheuristics such as Evolutionary Algorithms, Tabu Search and Simulated annealing have been widely studied, and have subsequently acheived relevance and success. Simulated annealing was inspired by classical mechanics, and was one of the earliest metaheuristics. The path-integral Monte Carlo version of quantum annealing is a population-based metaheuristic inspired by quantum mechanics.

It is demonstrated that quantum annealing is a highly competitive algorithm for graph coloring, which is an NP-hard problem having several applications including timetabling. In particular, quantum annealing is now the leading algorithm on the widely used DIMACS graph coloring benchmarks. Quantum annealing has found several new results such as 47-colorings for DSJC500.5, 82-colorings for SJC1000.5 and 400-colorings for C2000.9, none of which have yet to be reached by any
other algorithm.

SPEAKER’S BIO: Olawale Titiloye obtained a PhD from Manchester Metropolitan University in 2013, and a BSc in Information Systems and Management from Birkbeck College, University of London. The title of his PhD thesis is “Optimization by Quantum Annealing for the Graph Colouring Problem”. His research interests include stochastic local search and exact algorithms for combinatorial optimization problems such as the traveling salesman, vehicle routing, graph colouring and Boolean satisfiability.

Reference person: Sven Helmer

Victor Codina

TIME: Feb 13, 2013, 11:00-12:00

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Victor Codina, Technical University of Catalonia (UPC) – Barcelona

TALK TITLE: Semantically-Enhanced Pre-Filtering for Context-Aware Recommender Systems

TALK ABSTRACT: Context-aware recommender systems aim at outperforming traditional 2-dimensional recommenders by exploiting information about the context under which the users’ ratings are acquired. In this talk I will present a novel contextual pre-filtering approach that takes advantage of the semantic similarities among contextual situations. For assessing context similarity we have proposed a method, based only on the available users’ ratings, which estimates as similar two contextual situations that are influencing in a similar way the user’s rating behavior (implicit semantics). I will describe the different variants of the semantically-enhanced pre-filtering approach that we have developed and evaluated. Finally, I will talk about the experimental results of our performance comparison using several contextually-tagged ratings data sets, showing that the proposed pre-filtering approach outperforms state-of-the-art context-aware recommendation techniques.

SPEAKER’S BIO: Victor Codina is a PhD student in the Artificial Intelligence (AI) doctoral program at Universitat Politècnica de Catalunya – BarcelonaTech (UPC) in Barcelona (Spain). His research focuses on developing novel semantically-enhanced (context-aware) prediction models able to exploit the implicit semantics of item attributes as well as contextual conditions. In 2009, he received his MSc degree in Artificial Intelligence from UPC, with a thesis on Ontology-based Recommender Systems applied to an online tourism application. Previous to this, he was a researcher focused on Personalization technologies at TMT Factory, where he developed a recommender system for multimedia content that was patented in 2009 (US patent US20090100094).

REFERENCE PERSON: Francesco Ricci (http://www.inf.unibz.it/~ricci/)

Fabio Persia

TIME: Feb 4, 2013, 11:00-12:00

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Fabio Persia, DIS, University of Naples

TALK TITLE: Discovering the Top-k Unexplained Sequences in Time-Stamped Observation Data

TALK ABSTRACT: There are numerous applications where we want to discover unexpected activities in a sequence of time-stamped observation data—for instance, we may want to detect inexplicable events in transactions at a web site or in video surveillance of an airport tarmac. So far, we start with a known set A of activities (both innocuous and dangerous) that we wish to monitor. However, in addition, we wish to identify “unexplained” subsequences in a sequence of observations that are poorly explained by A (e.g., because they may contain occurrences of activities that have never been seen or anticipated before, i.e. they are not in A). We formally define the probability that a sequence of observations is unexplained (totally or partially) w.r.t. A. We develop efficient algorithms to identify the top-k Totally and Partially Unexplained Sequences w.r.t. A. These algorithms leverage a set of theorems that enable us to speed up the search for totally/partially unexplained sequences. We describe experiments using real-world datasets in the video and cyber security domains showing that our approach works well in practice in terms of both running time and accuracy

SPEAKER’S BIO: Fabio Persia received the Master degree in Computer Science and Engineering in 2009 from the University of Naples “Federico II”, Italy, where he is currently a third-year PhD student in Computer Science and Engineering under the supervision of Professor Antonio Picariello and Engineer Vincenzo Moscato. His present research interests lie in the field of video surveillance applications, multimedia databases, and knowledge representation and management. From November 2010 to April 2011, during his PhD, he visited the University of Maryland, College Park (USA), where he worked on finding “unexplained” activities in video under the supervision of Professor V.S. Subrahmanian.

REFERENCE PERSON: Francesco Ricci (http://www.inf.unibz.it/~ricci/)

Maciej Piernik

TIME: Jan 30, 2013, 11:00-12:00

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Maciej Piernik, Poznań University, Poland

TALK TITLE: Pattern-based XML mining

TALK ABSTRACT: Now that the use of XML is prevalent, methods for mining semi-structured documents have become even more important. It is vital to fully take advantage of the characteristics XML has to offer and improve the performance of document processing tasks. In this presentation, I will focus on XML clustering and classification. First, I will investigate the main challenges XML clustering implies and introduce a pattern-based framework, called XPattern, which tries to tackle these challenges. Next, I will present an instance of the XPattern framework and discuss the experimental evaluation of the proposed solution. Finally, I will focus on the XML classification problem and introduce an idea for a pattern-based solution inspired by the k-nearest neighbours algorithm, called k-nearest patterns. This method utilizes a pattern-document distance measure, called partial tree edit distance, which will be presented along with preliminary results and future research directions.

SPEAKER’S BIO: Maciej Piernik is research assistant in the Laboratory of Computer Science at the Institute of Computing Science, Poznań University of Technology. He received his master degree from Poznań University of Technology. His current research considers pattern-based XML mining.

REFERENCE PERSON: Mateusz Pawlik (http://www.inf.unibz.it/~pawlik/)

Shuaiqiang Wang

TIME: Jan 11, 2013, 11:00-12:00

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Shuaiqiang Wang, Shandong University of Finance and Economics, China

TALK TITLE: Ranking-Based Collaborative Filtering

TALK ABSTRACT: Collaborative filtering (CF) is an effective technique addressing the information overload problem. Ranking-based CF has demonstrated advantages in recommendation accuracy, being able to capture the preference similarity between users even if their rating scores differ significantly. In this study, we propose VSRank, a novel framework that seeks accuracy improvement of ranking-based CF through adaptation of the vector space model. In VSRank, we consider each user as a document and her pairwise relative preferences as terms. We then use a novel degree-specialty weighting scheme resembling TF-IDF to weight the terms. Our previous work belongs to memory-based CF algorithms. In this talk we also discuss our research plan for VSRank in the directions of the model-based CF algorithms.

SPEAKER’S BIO: Shuaiqiang Wang is currently an Associate Professor at Shandong University of Finance and Economics. He received B.Sc. and Ph.D. in Computer Science from Information Retrieval Lab, Shandong University, China, in 2004 and 2009 respectively. He was an exchange Ph.D. student at AOC group, Hong Kong Baptist University in 2009. He was a postdoctoral research associate at Data Mining Lab, Texas State University – San Marcos in 2010. His research interests include information retrieval, data mining, and machine learning. Shuaiqiang published more than 20 papers in prestigious international conferences and journals. He served as program committee member for CIKM 2012, and reviewers for IEEE Transactions on Evolutionary Computation (TEVC), Knowledge and Information Systems (KAIS) and Computers and Industrial Engineering (CAIE).

REFERENCE PERSON: Francesco Ricci (http://www.inf.unibz.it/~ricci/)

Anton Dignös

TIME: Dec 19, 2012, 16:00-17:30

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Anton Dignös, University of Zurich

TALK TITLE: Temporal Alignment

TALK ABSTRACT:
Time is present in almost all application domains, and many
applications have to store and manage time-varying data. Valid-time
databases aim to provide support for the management of data with
attached valid-time intervals. For the processing of such data an
essential part is the adjustment of the time intervals, and the
scaling of some attribute values, such as project budgets along with
the adjusted time intervals. This talk introduces an approach that
allows a traditional RDBMS to fully support the processing of interval
timestamped data, by solely integrating two primitives. Temporal
operations such as aggregation and joins, can then be systematically
reduced to traditional database operations with the help of this
primitives.

SPEAKER’S BIO:
Anton Dignös received a B.Sc. in Applied Computer Science and a
M.Sc. in Computer Science from the Free University of
Bozen-Bolzano. Since 2010 he is a PhD student at the Database Research
Group of the University of Zürich. His main research interest is
Temporal Databases with a focus on providing native database support
for interval-timestamped data.

REFERENCE PERSON: Johann Gamper (http://www.inf.unibz.it/~gamper/)

Johann-Christoph Freytag

TIME: Nov 26, 2012, 11:00

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Johann-Christoph Freytag, Humboldt-Universität zu Berlin, Germany

TALK TITLE: Stratosphere – Above the Clouds or Using Map/Reduce for Implementing Next-Generation Database Systems

TALK ABSTRACT:
Over the last five years the MapReduce compute paradigm has gained momentum in the database community as a platform for managing large volumes of data. At the same time it has evolved as a platform for implementing a new generation of database management systems (DBMSs).

This talk first introduces the DFG funded research project Stratosphere – Information Management above the Clouds, a project (Forschergruppe) among several database research groups in the Berlin area (see http://www.stratosphere.eu/). Within Stratosphere, the DBIS Research Group at the Humboldt-Universität zu Berlin focuses on adaptive query processing.

This talk also presents an overview on our first results on implementing adaptive query processing techniques into Stratosphere. We first discuss how to exchange different join strategies during the execution of the join and how to approach the problems during adaptation. Furthermore, we describe our operator competition model that allows the system to make decisions about the best operator (or partial query execution plan) “on the fly” after partially execution. We also demonstrate our approach by showing an experimental prototype system.

SPEAKER’S BIO:
Johann-Christoph Freytag is currently full professor for Databases and Information Systems (DBIS) at the Computer Science Department of the Humboldt-Universität zu Berlin, Germany. Before joining the department in 1994, he was a research staff member at the IBM Almaden Research Center (1985-1987), a researcher at the European Computer-Industry-Research Centre (ECRC, in Munich, Germany, 1987-1989), and the head of Digital’s Database Technology Center (also in Munich, 1990-1993). He holds a Ph.D. in Applied Mathematics/Computer Science from Harvard University, MA.

Prof. Freytag’s research interests include all aspects of query processing and query optimization in object-relational database systems, new developments in the database area (such as semi-structured data, data quality, databases and security), privacy in database systems, and applying database technology to applications such as GIS, genomics, and bioinformatics/life science. In the last years he received the IBM Faculty Award four times for collaborative work in the areas of databases, middleware, and bioinformatics/life science. He organized the VLDB conference in Berlin in 2003 and was a member of the VLDB Endowment (2001-2007) and in the head of the German database interest group of the GI (Fachbereich DBIS, Gesellschaft für Informatik).

REFERENCE PERSON: Johann Gamper (http://www.inf.unibz.it/~gamper/)

Jane Yao

TIME: Nov 19, 2012, 11:00

PLACE: DIS Public area, 2nd floor, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Jane Yau, Malmö University, Sweden

TALK TITLE: A mobile context-aware learning schedule framework with Java learning objects

TALK ABSTRACT:
In this seminar, I will present a theoretical mobile context-aware learning schedule framework which I had derived and designed from an extensive literature review, as part of my doctoral research. Its objective is to recommend appropriate learning materials to students based on their current locations and circumstances. The context of this research was on mobile learning i.e. learning in different locations and under various contextual situations from the perspective of university students. The framework uses a learning schedule (i.e. electronic-based diary) to inform the location and available time a student has for learning/studying at a particular location. Thereafter, a number of factors are taken into consideration for the recommendation of appropriate learning materials. These are the student’s learning styles, knowledge level, concentration level, frequency of interruption at that location and their available time for learning/studying. I conducted five studies to evaluate and determine the potential deployment of the framework and the results showed that (a) a learning schedule approach is successful to an extent in obtaining location and available time information to indicate accurate values of these contexts, (b) different learners may require different personalization strategies when selecting appropriate learning materials for them in mobile environments, (c) the proposed suggestion rules are effective in recommending appropriate materials to learners in their situation, in order to enhance their learning experiences, and overall that (d) the framework can potentially be used by students in different locations and situations, and appropriate learning materials can be selected to them, in order to enhance their learning experiences.

SPEAKER’S BIO:
Dr. Jane Yau is currently a Postdoc in Computer Science at Malmö University in Sweden, since June 2012. Prior to this, she received a Postdoctoral fellowship and joined the Center for Learning and Knowledge Technologies at Linnaeus University, Sweden, for 1.5 years. She completed her PhD in Computer Science specializing in the area of Mobile Learning in 2010 at the University of Warwick, UK. Her thesis was focused on a mobile context-aware learning schedule framework using Java learning objects. She has been/is currently involved in the following projects – mHealth, GEM (Geometry Mobile), LETS GO (learning about environment science), Co-create (collaborative story-telling with mobile devices), mobile services for energy efficiency, and Internet of Things. Her research interests include mobile learning, context-aware and recommender systems, mobile health and tourism. She has published around 30 scientific articles in the area of mobile learning.

REFERENCE PERSON: Francesco Ricci (http://www.inf.unibz.it/~ricci/)

Martin Theobald

TIME: Sep 26, 2012, 11:00

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Martin Theobald, Max Planck Institute for Informatics, Saarbrücken

TALK TITLE: Efficient Query Processing in Uncertain RDF Knowledge Bases

TALK ABSTRACT:
Recent advances of knowledge harvesting projects such as DBpedia and
YAGO have devised the way for the automatic construction and growth of
large, semantic knowledge bases from Web sources. The very nature of the
underlying extraction techniques however entails that the resulting RDF
knowledge bases may face a significant amount of incorrect, incomplete,
or even inconsistent (i.e., “uncertain”) factual knowledge. This talk
presents our results on query-driven reasoning techniques that address
the resolution of uncertainty within such data directly at query time.
Specifically, we present URDF, an efficient reasoning framework for
uncertain RDF knowledge bases. URDF provides a SPARQL-like query model,
and it combines rule-based, first-order predicate logic with
probabilistic inference techniques to derive new facts and to resolve
data uncertainty. The UViz visualization frontend dynamically accesses
and visually supports the URDF reasoning backend, thus providing an
intuitive user interface for exploring the knowledge base, visualizing
the steps involved in both the rule-based and probabilistic processing
steps, as well as explaining answers through lineage.

SPEAKER’S BIO:
Martin Theobald is a senior researcher at the Max Planck Institute for
Informatics. He obtained a doctoral degree in computer science from
Saarland University, and spent two years as a post-doc at Stanford
University where he worked on the Trio probabilistic database system.
Martin received an ACM SIGMOD dissertation award honorable mention in
2006 for his work on the TopX search engine for efficient ranked
retrieval of semistructured XML data. Martin is a member of the
advisory board of Elsevier’s Information Systems and served on the
program committee of many international journals, conferences and
workshops in the areas of Databases, Information Retrieval and the
Semantic Web, including CACM, ACM-TODS, IEEE-TKDE, IEEE-IS, PVLDB,
SIGMOD, SIGIR, ICDE and CIKM.

REFERENCE PERSON: Johann Gamper (http://www.inf.unibz.it/~gamper/)

Irena Koprinska

TIME: Sep 18, 2012, 11:00

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Irena Koprinska, University of Sydney

TALK TITLE: Recommender Systems for Online Dating

TALK ABSTRACT:
Online dating websites are used by millions of people. This talk will explore current research at the University of Sydney, Computer Human Adaptive Interaction (CHAI) research lab, to build recommender systems for online dating, in collaboration with a major Australian dating website. We will first show that similar people, as defined by a set of personal attributes, like and dislike similar people and are liked and disliked by similar people. This analysis provides the foundation for our content-collaborative recommender approach. We will then present a study of the implicit and explicit user preferences. The explicit preferences are stated by the user while the implicit preferences are inferred based on the user behavior on the website using a machine learning method. We will discuss the effectiveness of the implicit and explicit user preferences in predicting the success of user interactions.

SPEAKER’S BIO:
Irena Koprinska is a Senior Lecturer at the University of Sydney. She received PhD in Computer Science from the Institute for Information Technologies in Sofia, Bulgaria, in the area of Machine Learning, and Masters in Higher Education from the University of Sydney. Irena was a visiting researcher at the Technical University of Graz, Austria, and a post-doctoral fellow at the University of Trieste, Italy, and the University of Otago, New Zealand. Her research interests are in Machine Learning, Data Mining and Neural Networks, and their applications for Pattern Recognition, Recommender Systems and Personalisation. She regularly serves on the program committees of international conferences and also as a reviewer for journals and funding bodies.

REFERENCE PERSON: Francesco Ricci (http://www.inf.unibz.it/~ricci/)

Andrej Košir

TIME: Sep 5, 2012, 11:00

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Andrej Košir, University of Ljubljana

TALK TITLE: What can pattern recognition tells us about modelling recommender systems

TALK ABSTRACT:
The author will discuss the frequency domain aspect of user interaction in recommender system by observing it as different frequency bands (low- frequency stable behaviour to high frequency context-dependant behaviour). The author will then present a pattern recognition view of recommender systems where he will discuss the role of features, the between- and within-classes variance and the dynamical selection of variables to adopt in recommendations. The author will then proceed to the issue of selecting the relevant variables by applying a dynamic algorithm based on statistical testing and power analysis. From that point forward, the author will proceed on presenting the recommendation procedure as the prediction of human decision making. The author will present how the Ajzen model of decision making can be used for that scope. Among several decision making models, the Ajzen model has the nice property of not requiring to have all the features which fits well with the usual sparsity problem in recommender systems.

SPEAKER’S BIO:
Andrej Košir is associate professor at the Faculty of Electrical Engineering, University of Ljubljana. He was awarded the Vidmar prize for his educational prowess. He is active in several research fields, including signal, image and video processing, optimization (numerical optimization, genetic algorithm), and user interfaces. He was a guest researcher at the University of Westminster, London, UK, at the University of Waterloo, Canada, and at the North Carolina State University, USA. He is currently leading projects from the field of multimedia, optimization methods and digital signal processing – especially in object recognition on digital images, intelligent networks, user interfaces and user modeling.

REFERENCE PERSON: Francesco Ricci (http://www.inf.unibz.it/~ricci/)

Marko Tkalčič

TIME: Sep 4, 2012, 15:00

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Marko Tkalčič, University of Ljubljana

TALK TITLE: Affect in recommender systems

TALK ABSTRACT:
The author will focus on the usage of emotions in recommender system. The claim is that emotions do influence the performance of recommender systems. As it is known from other research areas, human decision making (e.g. choice of a film to watch) is influenced by emotions. Furthermore, users differ greatly in their emotion seeking tendencies (e.g. some people prefer horror films to light comedies). The author will present what emotions (and related mood and personalities) are, how we define them, how we elicit them and how we measure them. Then, the author will present a consumption chain-based model of the influence of emotions in different stages (decision, consumption, feedback) and give a historical overview of the usage of emotions in recommender systems. The author will focus on the usage of emotions in the decision making stage, where he will present the Kahneman/Tversky decision making model where two systems, an affective and a cognitive one, contribute to the final decision taken. The idea is that by including emotions in the KT model we can make more accurate predictions of human decision hence making better recommendations.

SPEAKER’S BIO:
Marko Tkalčič received his PhD degree at the University of Ljubljana Faculty of electrical engineering (UL FE). In 1999 he received the student Prešeren award for his BSc thesis. Since 1999 he has been employed as a researcher at the Digital signal processing Laboratory (LDOS) at UL FE where he has been working in various research areas including human visual perception, colour management, peer-to-peer networking, web interfaces and mobile services. In the years 2006-2007 he served as the technical manager of the FP6 eTEN P2PME project. In 2010 he founded the Affective Computing Students Interest Group (ACSIG) at UL FE. He is currently investigating other aspects of emotions and personality in human-computer interaction systems, especially in recommender systems.

REFERENCE PERSON: Francesco Ricci (http://www.inf.unibz.it/~ricci/)

Mohamed‐Amine Baazizi

TIME: Aug 23, 2012, 11:00

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Mohamed-Amine Baazizi, University of Paris Sud

TALK TITLE: Static Analysis for Optimizing the Update of Large Temporal XML Documents

TALK ABSTRACT:
The last decade has witnessed a rapid expansion of XML as a format for representing and exchanging data through the web. In order to follow this evolution, many languages have been proposed to query, update or transform XML documents. At the same time, a range set of systems allowing to store and process XML documents have been developed. Among these systems, main‐memory engines are lightweight systems that are the favored choice for applications that do not require complex functionalities of traditional DBMS such as transaction management and secondary storage indexes. These engines require to loading the documents to be processed entirely into main‐memory. Consequently, they suffer from space limitations and are not able to process quite large documents. XML projection is a prominent technique that has been proposed for overcoming the limitations of main‐memory engines in the context of querying. It consists in pruning out, at loading time, the parts of the queried document that are not required to perform the query. As in general queries are selective, the resulting document is likely to be small and fit within main‐memory. The projection technique developed for queries cannot be directly applied for updates. Obviously, updating a projection of a document t is not equivalent to updating the document t itself since the pruned out sub‐trees will be missing. Using projection in the context of updates requires thus a mechanism for propagating, on the original documents, the effects of the updates performed over the projected documents.

This talk consists in two parts. The first part focuses on presenting the projection‐based technique we have developed for optimizing the update of XML documents. This technique relies on a static analysis of the update expression and exploits the schema information. The conducted experiments testify the effectiveness of the method wrt space consumption and execution time as well.

The second part of my talk addresses the issue of efficiently building and maintaining timestamped XML documents encoding the evolution of XML documents. I will present two methods. For the first one, called general method, no restriction is made on the evolution of the XML documents whereas for the second one, called updatebased method, changes are assumed to be specified by updates. For both methods, the issue is to enable processing very large documents, to use existing engines and to comply to XQuery Update Facility. The two methods are compared in terms of space‐efficiency. The update‐based method produces time‐stamped XML documents that are more satisfactory wrt space‐efficiency than the general method.

SPEAKER’S BIO:
Mohamed‐Amine Baazizi is a PhD candidate at University of Paris Sud and INRIA Saclay in France. His research focuses on XML update optimization and on the management of temporal XML. The common basis of his research interests lay on the efficient processing of large XML data exploiting information about schemas and on the management of change‐centric XML data. Mohamed‐Amine received his master degree from University of Lyon in 2008 and engineering diploma from University of Technology of Algiers.

REFERENCE PERSON: Johann Gamper (http://www.inf.unibz.it/~gamper/)

Adolfo Urrutia

TIME: July 26, 2012, 11:00am-12:00pm

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Adolfo Urrutia, Technical University of Madrid

TALK TITLE: Outlining recommendation zones for taxi drivers using self-organizing maps

TALK ABSTRACT:
After a taxi driver dropped-off his last passenger, the search for the next customer begins. This search involves a number of steps like: Data cleansing, passenger pick-up zones creation, probability of finding a passenger, a recommendation model, rate of success of each recommendation, to mention some of them.

In this paper, using three of the previous mentioned steps, data from 50 Spanish taxicabs is analyzed. The data is from Madrid and it was capture during 45 days, giving a total of 3 millions of raw GPS-registers and other events. After a cleaning process, the taxi data was spatiotemporal clustered using a Self Organizing Map, in order to create the passenger pick-up zones. These zones change over time following a weekly and 24-hour behavior that reflect how the taxi demand fluctuates over Madrid. Due in a clustering process, is not possible to know, beforehand, how many clusters will be necessary, and because, as far as my knowledge, there is no other research about taxi spatiotemporal zones in Madrid, it was decided to do a comparative study. This research shows how the amount of clusters, temporal representation and data normalization affect the spatiotemporal size of the passenger pick zones, and its probability for providing new passengers.

SPEAKER’S BIO:
Adolfo Urrutia obtained his Bsc in computer Engineering from the National University of Engineering, Nicaragua. His final engineering project was a GIS-based software tool for the calculation of seismic risk. Since 2009 is a Phd Student at Technical University of Madrid, focusing in spatiotemporal datamining, machine learning techniques and recommendation systems. He has been involved in different research projects, one of them Dycoopnet (Dynamic Complexity of Cooperation-Based Self-Organizing Networks in the First Global Age). His current research line is focused on network-constrained moving objects behavior: Study case GPS-enabled taxicabs for recommendation of pick-up zones.

REFERENCE PERSON: Johann Gamper (http://www.inf.unibz.it/~gamper/)

Victor Codina

TIME: July 10, 2012, 11:00-12:00

PLACE: Seminar room POS 1.02, Dominikanerplatz 3, 39100 Bolzano

SPEAKER: Victor Codina

TALK TITLE: Semantically-Enhanced, Context-Aware Recommendation Algorithms

TALK ABSTRACT:
Recommendation models based on collaborative filtering have recently gained much popularity because of their good prediction accuracy. However, collaborative-filtering models have very poor performance in cold-start scenarios or in “dynamic” domains, when no or few rating data are associated to each item. This limitation is even more aggravated in some context-aware models because only rating data associated with the active context is used for recommendation. In these situations content-based filtering models usually work better, but traditional approaches suffer from two main limitations: (1) the limited quality and quantity of available item metadata; (2) the lack of understanding and exploitation of available domain semantics. Precisely, this second issue is what this research focuses on and where we aim to take a step forward with respect to previous work. Approaches dealing with this issue are known as semantically-enhanced recommenders and consist of knowledge-based hybrids that exploit semantic relations among item attributes to enhance the recommender performance. In this talk I will present the semantically-enhanced, context-aware models which I have evaluated so far and discuss the main results with two different data sets. The main novelty of this work with respect to the state of the art is that the models developed can be applicable to any kind of semantic relation between item attributes or contextual conditions, and not only to taxonomic relations, as seen in most previous work. In this way, the semantically-enhanced models are not limited to the availability of explicit domain/contextual knowledge in the form of ontologies, which is usually difficult and expensive to obtain.

SPEAKER’S BIO:
Victor Codina is a PhD student in the Artificial Intelligence PhD program at Technical University of Catalonia (UPC), Spain. He obtained his bachelor degree in Computer Science from UPC in 2007, where he was working on the development of a hybrid recommender for an interactive television system provided by the company TMT Factory. In 2009, he received his MSc degree from UPC, where he developed a travel recommender using as main recommendation strategy a semantically-enhanced model that exploited the taxonomical relations between item attributes defined in several domain ontologies. Currently, he is working in his PhD thesis whose proposal was accepted in 2011 under the supervision of Dr. Luigi Ceccaroni. His research focuses on developing new semantically-enhanced models that can take advantage of any type of semantic relations between item attributes as well as contextual conditions (in context-aware settings).

REFERENCE PERSON: Francesco Ricci (http://www.inf.unibz.it/~ricci/)

Theodoros Chondrogiannis

TIME: July 4, 2012, 11:30-12:30

PLACE: Seminar room POS 1.02, Dominikanerplatz, 39100 Bolzano

SPEAKER: Theodoros Chondrogiannis, Department of Informatics and
Telecommunications at the National and Kapodistrian University of Athens (NKUA), Greece

TALK TITLE: A Geography-aware Service Overlay Network for Managing Moving Objects

TALK ABSTRACT:
As the proliferation of mobile devices and positioning systems
continues unabated, the need to provide more robust
location-based services becomes more pressing. In this context,
we examine the problem of efficiently handling queries over
moving objects and propose a location-aware overlay network that
helps monitoring such objects while traversing contained
geographic extends. We use a triangulation structure to divide a
geographic area using fixed service nodes as anchors based on
their geographic position. Triangulation inherently contains each
moving object within an area designated by three service
nodes. We introduce a method for monitoring moving objects and we
present an algorithm for processing nearest-neighbor queries
while restricting the amount of resources and, subsequently, the
volume of transmitted messages. Through simulation, we evaluate
the suggested approach and show that our nearest-neighbor query
processing method provides always accurate results while it uses
invariantly a constant number of service nodes. We finally show
that the average physical distance between service and roaming
nodes remains limited; this yields a significant number of
physical connections that avoid conventional Internet routing
altogether.

SPEAKER’S BIO:
I am an MSc student at the Department of Informatics and
Telecommunications at the National and Kapodistrian University of
Athens in Athens (NKUA), Greece. I received my 4-year Bachelor
degree in Computer Science and Technology from the University of
Peloponnisos in Tripolis, Greece. During my studentship at the
University of Peloponnisos, I conducted research on the
plausibility of cardinal directions relations models, under the
co-supervision of Dr. Spiros Skiadopoulos and Dr. George
Lepouras. I currently work on my master thesis under the
supervision of Dr. Alex Delis. My research is focused on
location-based services and, especially, location-aware mobile
device monitoring.

REFERENCE PERSON: Johann Gamper (http://www.inf.unibz.it/~gamper/)

Boris Glavic

TIME: Feb 7, 2012, 11:00-12:00

PLACE: Seminar room POS 1.02, Dominikanerplatz, 39100 Bolzano

SPEAKER: Boris Glavic, University of Toronto, Canada

 

TALK TITLE: Provenance for Databases and Data Integration

TALK ABSTRACT: Current data management technologies including scientific databases, data warehouses, data integration frameworks, web technologies, and workflow management systems have enabled the recording and rapid sharing of enormous amounts of information. A large portion of such data is no longer the direct result of measurements or manually created by a user, but rather derived from existing data through complex automated transformations. In such settings it is of utmost importance to understand the origin and creation process of data to estimate its quality, to gain additional insights about it, or to trace errors in transformed data back to its origins. This kind of information is often referred to as data provenance.

In this talk I give an overview of how I address these challenges in my work in database systems. I will focus mainly on two of my projects in data provenance: (1) Perm (http://permdbms.sourceforge.net/) is a scalable system for generation and querying of provenance information over relational data. The two key ideas behind this approach are representing data and its provenance together in a single relation and rewriting queries to generate this representation. Perm supports fully integrated, efficient, on-demand provenance generation and querying. (2) The TRAMP system enables debugging of information integration scenarios based on provenance information. The system supports tracing errors with different types of causes (the data, inconsistencies between data sources, the schemas, schema constraints, the mappings, or the transformations). TRAMP combines data provenance with two novel notions, transformation provenance and mapping provenance, to explain the relationship between some transformed data and those transformations and mappings that produced that data.

SPEAKERS HOMEPAGE: http://www.cs.utoronto.ca/~glavic/

SPEAKER’S BIO:
My research interest is in the field of database systems, in particular, data provenance and its application in information integration. I finished my Diploma in Computer Science at RWTH Aachen, Germany in 2005 and my Ph.D. in Computer Science at the University of Zurich, Switzerland in 2010. I am currently working as a postdoctoral Fellow at University of Toronto being advised by Renee J. Miller.

In my research I focus on building complete systems that automatically and efficiently generate provenance information and present such information to the user in a format that is easily comprehensible and query-able. In my Ph.D. thesis project Perm, I have extended the PostgreSQL database system to support generation and querying of provenance for SQL queries. Furthermore, I have developed Ariadne, a data stream management system that natively generates provenance for streaming queries. In addition to building provenance-enabled systems, I have investigated how other areas of data management can benefit from and be built on-top of such systems. For instance, the TRAMP and Vagabond approaches I developed have enabled debugging of information integration scenarios based on provenance information.

 

I have served as a program committee member for SIGMOD 2011 and the AWM workshop 2012, as a reviewer for TODS and TKDE, and as an external reviewer for ICDE 2008 and 2010, VLDB 2006 and 2007, SIGMOD 2007, TOOLS 2007, BIS 2007, and BTW 2009.

REFERENCE PERSON: Periklis Andritsos (http://www.cs.toronto.edu/~periklis/)

 

Arturas Mazeika

TIME: Jan 27, 2012, 15:00-16:00

PLACE: Seminar room POS 1.02, Dominikanerplatz, 39100 Bolzano

SPEAKER: Arturas Mazeika, MPI Saarbrücken, Germany

 

TALK TITLE: Visual Analytics of Evolving Named Entities

SPEAKERS HOMEPAGE: http://www.mpi-inf.mpg.de/~amazeika/

SPEAKER’S BIO:
Arturas Mazeika is a researcher in the Database and Information Systems Group at Max-Planck-Institut für Informatik, Germany. He holds a MSc from the Department of Mathematics and Computer Science at Vilnius University, and PhD at the Departments of Computer Science and Communication at Aalborg University. His research interests include Information retrieval, visual data mining, data mining, and databases.

REFERENCE PERSON: Johann Gamper (http://www.inf.unibz.it/~gamper/)

TALK ABSTRACT: The constantly evolving Web reflects the evolution of society. Knowledge about entities (people, companies, political parties, etc.) evolves over time. Facts add up (e.g., awards, lawsuits, divorces), change (e.g., spouses, CEOs, political positions), and even cease to exist (e.g.,
countries split into smaller or join into bigger ones). Analytics of the evolution of the entities poses many challenges including extraction, disambiguation, and canonization of entities from large text collections as well as introduction of specific analysis and interactivity methods for the evolving entity data. The presentation will cover the issues to prepare the data as well as build an analytic system to browse and analyze the evolution.

Anton Dignös

TIME: Dec 1, 2011, 14:00-15:00

PLACE: Seminar room POS 1.02, Dominikanerplatz, 39100 Bolzano

SPEAKER: Anton Dignös, IFI, University of Zurich

 

TALK TITLE: Curbing Time Intervals in Database Systems

SPEAKERS HOMEPAGE: http://www.ifi.uzh.ch/databaseTechnology/Staff/dignoes.html

SPEAKER’S BIO:
Anton Dignös received a B.Sc. in Applied Computer Science and a M.Sc. in Computer Science from the Free University of Bozen-Bolzano. Since 2010 he is a PhD student at the Database Research Group of the University of Zürich. His main research interest is Temporal Databases with a focus on providing native database support for interval-timestamped data.

TALK ABSTRACT:

Time is present in almost all application domains, and many applications have to store and manage time-varying data. Temporal databases aim to provide specific support for the management of such data. A lot of research has been conducted in this field over the past decades, focusing mainly on data representation, data models, query languages, indexing, and efficient evaluation algorithms for specific operators. There is little work about integrating temporal support in a DBMS in a principled way.
This talk is about an approach to provide native support for interval-timstamped data in a RDBMS. The talk will focus on the formal aspects of the approach as well as on its implementation in the kernel of a DBMS.

REFERENCE PERSON: Johann Gamper (http://www.inf.unibz.it/~gamper/)

 


Markus Schedl

TIME: November 21, 2011, 11AM

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Markus Schedl, Department of Computational Perception, Johannes Kepler University, Linz, Austria

TALK TITLE: Inferring Music Similarity from Content and Context

SPEAKER’S HOMEPAGE: http://www.cp.jku.at/people/schedl/

SPEAKER’S BIO:
Markus Schedl graduated in Computer Science from the Vienna University of Technology. He earned his Ph.D. in Computational Perception from the Johannes Kepler University Linz, where he is employed as assistant professor at the Department of Computational Perception. He further holds a Master’s degree in International Business Administration from the Vienna University of Economics and Business Administration. Markus (co-)authored more than 50 refereed conference papers and journal articles. Furthermore, he serves on various program committees and reviewed submissions to several conferences and journals. His main research interests include music and multimedia information retrieval, social media mining, information visualization, and intelligent user interfaces. He is co-founder of the “International Workshop on Advances in Music Information Research” and co-organizer of the “3rd International Workshop on Search and Mining User-generated Contents” and the “8th International Workshop on Adaptive Multimedia Retrieval”.

TALK ABSTRACT:
In this talk I will give an overview of computational methods to estimate the similarity between music items, on the artist-level and track-level. Traditionally, such similarity estimation approaches rely on audio signal processing techniques to derive features to which similarity measures are applied. These content-based techniques are well-suited to describe musical similarity on the track-level in terms of rhythm or timbre, for example. They fail, however, to take into account factors that are not encoded in the audio signal, nevertheless influence the perceived similarity, such as the lyrics’ meaning or the performer’s political or cultural background. Given that huge amounts of data describing this “music context” are becoming more and more available (e.g., collaborative tags, Web pages about artists, microblogs about listening activities), several contextual music similarity measurement approaches have been proposed in recent years. I will summarize the state-of-the-art in content- and context-based music similarity computation and quickly outline some aspects beyond pure similarity, which I deem to be important for personalized music retrieval.

 

 

REFERENCE PERSON: Francesco Ricci (http://www.inf.unibz.it/~ricci/)


Wolfgang Nejdl

TIME: July 8 at 11AM

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Wolfgang Nejdl, L3S / Leibniz Universität Hannover, Germany

TALK TITLE: Web Science @ L3S – Interdisciplinary Research Challenges

SPEAKER’S HOMEPAGE: http://www.kbs.uni-hannover.de/~nejdl/

SPEAKER’S BIO:
Prof. Dr. Wolfgang Nejdl, geboren 1960 in Wien, hat seit 1995 eine
C4-Professur für Informatik an der Leibniz Universität Hannover inne. Er
studierte Informatik an der TU Wien, die er 1984 mit dem Dipl.Ing. und
1988 mit dem Dr.techn. abschloss. 1992 habilitierte er, ebenfalls an der
TU Wien, und war von 1992 – 1995 C3-Professor an der RWTH Aaachen. Er
arbeitete als Gastwissenschaftlicher und –professor am Xerox PARC,
Stanford University, der University of Illinois at Urbana-Champaign,
EPFL Lausanne, and am PUC Rio.

Prof. Nejdl leitet seit 2001 das Forschungszentrum L3S und forscht in
den Bereichen Suche und Information Retrieval, Informationssysteme und
Datenbanken, Semantic Web und computerunterstütztem Lernen. Aktuelle
Projekte im L3S sind unter anderem Projekte zur Verbesserung der Suche
im Web, etwa das Projekt LivingKnowledge, das sich mit Diversität und
Meinungen im Web beschäftigt und das Projekt GLOCAL, das eventbasierte
Suche für multimediale Daten im Web enwickelt. Er veröffentlichte
bisher mehr als 260 wissenschaftliche Artikel, gelistet auf DBLP, und
war Programmkomitee-Vorsitzender, Programmkomitee-Mitglied und
Editorial-Board-Mitglied zahlreicher internationaler wissenschaftlicher
Konferenzen und Zeitschriften, wie zum Beispiel der ACM Transactions on
the Web oder als Vorsitzender das ACM Conference on Web Search and Data
Mining 2011 und der IEEE International Conference on Data Engineering
2011.

TALK ABSTRACT:

In the past 17 years, the Web has developed into a worldwide
information and communication infrastructure with a considerable
influence on business, science, and society. Despite its unprecedented
growth, or perhaps because of it, and because of the wealth of
applications that build on this infrastructure, the challenges faced
by the Web as a whole are also greather than they were 17 years
ago. Many of the necessary solutions can only be developed with a
combination of methods and technologies taken from various areas of
information technology and computer science – supported by
contributions from social sciences, business, and law.

In this talk I will give an overview over the current L3S core
research areas and some of our results, as well as on our planned
extension beyond computer science with colleagues from sociology,
psychology, linguistics and other disciplines.

REFERENCE PERSON Johann Gamper: gamper@inf.unibz.it


Christian Koncilia

TIME: June 21, 2PM-3PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Christian Koncilia, University of Klagenfurt, Austria

TALK TITLE:

Advanced Data Warehouse Analysis

SPEAKER’S HOMEPAGE:

http://www.xing.com/profile/Christian_Koncilia

SPEAKER’S BIO:

Dr. Christian Koncilia works as a PostDoc at the University of Klagenfurt, Austria. He received his MSc and his PhD in Computer Science with publications in the area of Data Warehouse and OLAP systems. Before joining the University of Klagenfurt in 2009 he worked as Managing Consultant for a data warehouse vendor in Munich, Germany and as IT Project Manager for an Austrian Hospital.

TALK ABSTRACT:

Nowadays, Data Warehouse systems and Business Intelligence tools are widely spread in all industries. They enable users to performantly analyse mass data by storing data in Non-SQL database systems or traditional relational database systems. Although nearly all data warehouses store sequential and temporal data, i.e. data with a logical or temporal ordering, traditional data warehouse or OLAP approaches fail when it comes to analyze those data. In this talk, I will present an approach to deal with temporal data and an approach that enables the user to analyse sequential data.

REFERENCE PERSON: Johann Gamper , gamper@ing.unibz.it


TIME: July 20, 11AM-12PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Alexandros Karatzoglou, Telefonica Research in Barcelona, Spain

TALK TITLE:

Collaborative Ranking for Recommender Systems

SPEAKER’S HOMEPAGE:

http://www.ci.tuwien.ac.at/~alexis/

SPEAKER’S BIO:

Alexandros Karatzoglou is currently a researcher at Telefonica Research in Barcelona, Spain. His research interests include Recommendation, Context, User Modeling and Machine Learning. Alexandros completed his PhD studies in 2006 in the field of Machine Learning and in particular kernel-based methods at the Vienna University of Technology. He then joined the Statistical Machine Learning Group at National ICT Australia as a visiting scholar where he started working on the field of Recommender Systems and Collaborative Filtering. In 2008 Alexandros joined the Machine Learning LITIS lab at INSA de Rouen in France, where he continued his work on Recommender Systems. Alexandros is the author of the popular kernlab R package for kernel-based Machine Learning. His work on Collaborative Filtering and Ranking was awarded the best Machine Learning paper award at ECML/PKDD 2008 and resulted in a software implementation called CoFiRank available at http://cofirank.org.

TALK ABSTRACT:

In this talk we are going to present collaborative ranking methods that are based on Matrix Factorization. In contrast to traditional Collaborative Filtering methods that aim at modeling the rating that a user might give to an item Ranking based methods aim at modeling the order in which the items are preferred by the user. The Collaborative Ranking models we present are based on a number of ranking loss functions and present some unique challenges.We employ state-of-the-art Machine Learning methods to deal with these challenges. We are also going to discuss optimization methods and implementation details.

REFERENCE PERSON: Francesco Ricci , Francesco.Ricci@unibz.it


TIME: July 1st, 11AM-12PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Neal Lathia, University College London

TALK TITLE:

Mobile Recommendations for Smart Cities

SPEAKER’S HOMEPAGE:

http://www.cs.ucl.ac.uk/staff/n.lathia/

SPEAKER’S BIO:

Neal is a post-doctoral researcher in the Software Systems Engineering Group of the Department of computer science, University College London. He is working on the EC i-Tour (Intelligent Transport Systems for Optimized Urban Trips) project, and does research on the intersection of mobility and personalization. He recently defended his PhD dissertation, entitled “Evaluating Collaborative Filtering Over Time,” also from UCL – supervised by Prof. Steve Hailes and Dr. Licia Capra.

TALK ABSTRACT:

Cities have become the heart of modern life and home to more than half of the world’s population: a key challenge of the future is developing solutions to the many complex problems related to urban environments. Recommender systems, which (to date) have mainly been implemented online, may be a key technology to address this scenario. In this talk, I’ll give an overview of how mobiles and passive sensors in the city can be used to build personalised services. In particular, we have been focusing on two contexts: (1) recommending social and cultural events by using location data, and (2) personalised travel information services for the city’s dynamic public transport infrastructure based on sensor data.

REFERENCE PERSON: Francesco Ricci , Francesco.Ricci@unibz.it


TIME: June 1st, 11AM-12PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKERS: Claudio Eccher and Chiara Ghidini(Fondazione Bruno Kessler – irst, Trento, Italy)

TALK TITLE:

Modeling the information flow process in hospitals: an experience in an oncology ward”

SPEAKER’S BIO:

Dr. Claudio Eccher graduated cum laude in physics in 1996 at the

University of Trento, Italy, with a thesis on the analysis and

classification of breast cancer images for the determination of tumour

angiogenesis. In 2006 he received the Ph.D. degree in Communication

and Information Technology from the ICT International Doctorate

School, University of Trento, working on the automatic translation of

Systems Biology Mark-Up Language (SBML) models of network of

biochemical reactions into the Biochemical Stochastic pi-calculus.

Researcher at the eHealth Unit of the Bruno Kessler Foundation

(formerly Istituto Trentino di Cultura), Center for Scientific and

Technological Research, Trento, Italy since 1997. He has worked in the

fields of Bioinformatics (translation of SBML into process algebra)

and Medical Informatics (image analysis in medicine, robotic tele-

pathology systems for intra-operative remote diagnosis, web-based

clinical records and tools for shared management of chronic patients,

use of medical ontologies for clinical system interoperability,

guideline-based decision support systems, business process modelling

applied to careflows).

He is co-author of several scientific papers in the areas of image

analysis in medicine, computer-based systems supporting shared patient

care, modelling languages for systems biology and the use of

ontologies for designing electronic medical records.

Dr. Chiara Ghidini is a Senior Research Scientist in the Data &

Knowledge Management unit at FBK. Dr. Ghidini completed her PhD in

computer science Engineering in 1998 at the University of Rome “La

Sapienza”. Before joining FBK she has worked as a post-doc at the

Centre for Agent Research and Development, Manchester Metropolitan

University, and as a lecturer at the Department of computer science,

University of Liverpool. Her work in the area of distributed knowledge

representation is well known and internationally recognized and she

has published over 50 conference and journal papers on the topics. She

has served as PC member in some of the most important conferences in

the area of the semantic web and multi-agent systems, as general co-

chair of the 2nd European workshop on multi-agent systems (EUMAS

2004), as programme co-chair of the 4th International and

Interdisciplinary Conference on Modeling and Using Context (CONTEXT

2003), and as tutorial co-chair of the 6th Annual European Semantic

Web Conference (ESWC 2009). She has been involved both in national and

international research projects, including the EU-funded FP6 APOSDLE

Project and the FP7 founded Organic.Lingua Project.

Enrico Maria Piras is a researcher at Fondazione Bruno Kessler. He is a sociologist and holds a PhD in “Information systems and Organization”. His primary research interests are related to the social and organizational issues related to the infrastructuring in the healthcare sector with particular regard to the information systems aimed at professionals and laypeople.

He works primarily in the areas of computer Supported Cooperative Work, organizational theory and qualitative methods in information systems research (ethnography and in-depth interviewing). He is also a member of the university-based Research Unit on Communication, Organizational Learning and Aesthetics of the Trento University (Rucola; http://www3.unitn.it/rucola/) a group of scholars and researchers collaborating since 1993 on the basis of common professional interest in specific aspects of Organization studies.

TALK ABSTRACT:

In hospitals, care delivery relies on information flowing

through several information systems (both electronic and paper-based).

The management of this information flow in different departments is

subject to various policies and rules, leading organizational actors

to have only a limited understanding of the whole process of care

delivery. BPMN (Business Process Modeling Notation) is a standard for

process modelling able to represent complex processes yet providing a

graphical notation intuitive to domain experts. Still, eliciting the

information management practices in order to create the BPMN models is

a difficult and complex task because these practices take place in a

chaotic environment that requires subtle adaptation to the situation

at hand. Moreover, communication often relies on taken-for-granted

activities that organization actors carry out off the top of their

heads, hence difficult to recall to memory for the benefit of the

analysts.

In this seminar we present a methodology that we have followed to

formalize in BPMN the information management practices performed by

nurses in an oncology department. The modeling was performed by an

interdisciplinary research team, following a process based on the

iteration between the “participant observation” in the department

(ethnographer), team meetings to analyse the observations and produce

coarse models (ethnographer, computer scientists), and modelling

activity (computer scientist) to refine the model with a dedicated

tool (Moki). Exploiting the richness of the observations and the

versatility of the tool, the process led to a set of BPMN models of

nurse information management in the oncological department that allows

describing the flow of the activities, the tools needed to carry them

out, and the roles played by nurses in the process.

REFERENCE PERSON: Floriano Zini , floriano.zini@unibz.it


TIME: June 1st, 10AM-11AM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Floriana Grasso , University of Liverpool

TALK TITLE:

Computational Argumentation for Digital Interventions

SPEAKER’S HOMEPAGE:

http://www.csc.liv.ac.uk/~floriana/Home.html

SPEAKER’S BIO:

Floriana Grasso is a Lecturer in the Department of computer science at the University of Liverpool. She works in the recent research field of Argument and Computation, where she is interested in persuasiveness, and what it takes to achieve effective (and affective) communication. She approaches this from two points of view: high level discourse modelling and rhetoric (modelling the speaker’s goals and strategies to generate persuasive discourse), where she draws on classical argumentation theory, and cognitive modelling, especially on modelling extra-rational characteristics such as opinions, values, emotions. She applied this research in public health informatics, where she especially looks at how to provide personalised and persuasive advice on healthier lifestyles. She has organised many workshops and events in the area of argumentation, persuasive technology, personalised eHealth, and motivation. She is co-editor of the journal “Argument and Computation”, published by Taylor and Francis.

TALK ABSTRACT:

Over the past decade or so, a new interdisciplinary field has emerged as a core and autonomous discipline within AI, starting at the intersection of artificial intelligence and the area of philosophy concentrating on the language and structure of argument, and positioning itself to cover knowledge representation and reasoning, as well as linguistics and cognitive science. An increasing number of venues, conferences, events, and now a scholarly journal, are now dedicated to the area of “Argument and Computation”. Applications have proliferated, in law, e-government, e-commerce, recommender systems, focus group analysis, public health informatics, and many more. In this talk I will start with a brief overview of this research area, and then move to show how insights from this field are useful when modelling motivational systems, in the scope of digital interventions for encouraging behaviour change.

REFERENCE PERSON: Floriano Zini , floriano.zini@unibz.it


TIME: May 26, 3PM-4PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Periklis Andritsos (the University of Toronto and University Health Network)

TALK TITLE:

What can clusters reveal about your data: sieving through unstructured information sources

SPEAKER’S BIO:

Periklis Andritsos received his B.Sc. degree in Electrical and computer Engineering from the National Technical University of Athens, Greece. He holds an M.Sc. and Ph.D. degree in computer science from the University of Toronto, where he also spent a year as a Post-doctoral fellow. From 2005 until 2008, he was an assistant professor at the Department of Information Engineering and computer science of the University of Trento, where he was a member of the Data and Knowledge Management group and the co-ordinator of the Bolzano-Innsbruck-Trento school. He has supervised a number of graduate students and his interests include database systems, data and text mining, clustering and reverse engineering. He joined Thoora in July of 2008, where he led the Research team and its efforts in solving real-life and large-scale industrial research problems. He is currently a Senior Researcher at the University of Toronto and University Health Network. He is a senior member of the IEEE computer Society and the Association for Computing Machinery.

TALK ABSTRACT:

Clustering is a problem of great practical importance in numerous applications. The problem of clustering becomes more challenging when the data is categorical, that is, when there is no inherent distance measure between data values. In this talk, we describe a scalable hierarchical categorical clustering algorithm and talk about the use of information loss for the identification of duplication in the data and finally present two real case studies : one in which categorical clustering is being used for the quantitative analysis of news sources and a second in which we leverage cluster information to understand how we can extract attribute dictionaries from unstructured text.

REFERENCE PERSON: Johann Gamper , gamper@inf.unibz.it


TIME: May 25, 3PM-4PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Benjamin Gufler, TU Munich

TALK TITLE:

Handling data skew in MapReduce

SPEAKER’S BIO:

Benjamin Gufler is a Ph.D. student at the Database Systems Group in the Department of computer science at TU München, Germany. He received his university diploma in computer science, with focus on database systems, from TU München in 2006. His research interests include cloud and grid

computing, especially for e-science applications, as well as load balancing and data mining.

TALK ABSTRACT:

MapReduce systems have become popular for processing large data sets

and are increasingly being used in e-science applications. In contrast

to simple application scenarios like word count, e-science

applications involve complex computations which pose new challenges to

MapReduce systems. In particular, (a) the runtime complexity of the

reducer task is typically high, and (b) scientific data is often

skewed. This leads to highly varying execution times for the

reducers. Varying execution times result in low resource utilisation

and high overall execution time since the next MapReduce cycle can

only start after all reducers are done.

In this paper we address the problem of efficiently processing

MapReduce jobs with complex reducer tasks over skewed data. We define

a new cost model that takes into account non-linear reducer tasks and

we provide an algorithm to estimate the cost in a distributed

environment. We propose two load balancing approaches, fine

partitioning and dynamic fragmentation, that are based on our cost

model and can deal with both skewed data and complex reduce

tasks. Fine partitioning produces a fixed number of data partitions,

dynamic fragmentation dynamically splits large partitions into smaller

portions and replicates data if necessary. Our approaches can be

seamlessly integrated into existing MapReduce systems like Hadoop. We

empirically evaluate our solution on both synthetic data and real data

REFERENCE PERSON: Nikolaus Augsten , augsten@inf.unibz.it


TIME: May 19, 3PM-4PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Christian von de Weth (Karlsruhe Institute of Technology (KIT), Germany)

TALK TITLE:

FAST: Friends Augmented Search Techniques – System Design & Data-Management Issues

SPEAKER’S BIO:

Christian von der Weth got his university diploma in computer science – with focus on database and information systems – in 2004 from the Ilmenau University of Technology (Germany). He, then, started as research fellow at the Karlsruhe Institute of Technology (KIT, Germany). In 2009 he finished his Ph.D.; his thesis addresses trust management and the stimulation of cooperative behaviour in virtual communities. Between January 2010 and February 2011 he worked as PostDoc in the School of computer Engineering at the Nanyang Technological University (Singapore). The goal of the project was a system for the support of collaborative browsing and searching the Web. Christian’s researches and interests range from trust management, online reputation systems, establishing cooperation in anonymous environments, social network analysis and behavioural economics, to data management and relational query processing in structured peer-to-peer networks. He served as reviewer for various computer science journals and conferences and is currently Deputy Information Director of the VLDB Journal.

TALK ABSTRACT:

Improving web search solely based on algorithmic refinements has reached a plateau. The emerging generation of searching techniques attempts to harness the “wisdom of crowds”, using inputs from end users in the spirit of Web 2.0. This talk introduces a framework facilitating friends augmented search techniques (FAST), with focus on system design & data-management issues. It first presents a browser add-on as front end for collaborative browsing and searching, supporting both synchronous and asynchronous collaboration between users. The main part of the talk addresses the back end system supporting data-management for FAST, a distributed key-value store for an efficient information retrieval in the presence of an evolving knowledge base. The developed indexing and query processing techniques are of more general interest and applicability.

REFERENCE PERSON: Mouna Kacimi , Mouna.Kacimi@unibz.it


TIME: May 18, 3PM-4PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Saad Malik (University of Toulouse, France)

TALK TITLE:

Combining Granularity-Based Topic-Dependent and Topic-Independent Evidences for Opinion Detection

SPEAKER’S BIO:

Malik is a doctoral student in IRIT (Institut tde Recherch en Informatique de Toulouse), France. Opinion Detection is his main area of interest of his PhD studies. During his career he has been associated with the field of Usability and human-computer Interaction. He holds two masters degrees with specializations in Usability and Human-computer Interaction (Pakistan) and Information, Image and Hypermedia (Toulouse, France). Mr. Malik has worked under supervision of many visionary IR researchers from various famous IR labs (like Yahoo! Research Labs, L3S ). After successfuly defending his work in many International conferences, Mr. Malik will defend his thesis in June, 2011.

TALK ABSTRACT:

Opinion mining is a subdiscipline within Information Retrieval (IR) and Computational Linguistics. It refers to the computational techniques for extracting, classifying, understanding, and assessing the opinions expressed in various online sources like news articles, social media comments, and other user-generated content. Like any other research field, there are many challenges associated with the field of opinion mining. In this talk, I will focus on some major challenges of opinion mining and the way we and research community is dealing with it. Our contributions in this regard deal with the problems of

1) finding opinion-topic associations within opinionated documents, ,

2) combining topic-dependent and topic-dependent evidences for opinion detection,

3) paving ways for entity-based opinion detection,

4) using social network based evidences for opinion detection

REFERENCE PERSON: Mouna Kacimi , Mouna.Kacimi@unibz.it


TIME: January 21, 11AM-12PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Giovanni Semeraro (University of Bari “Aldo
 Moro”, Italy)

TALK TITLE:

Content-based Recommender Systems: problems, challenges and research directions

SPEAKER’S HOMEPAGE:

http://lacam.di.uniba.it:8000/people/semeraro.htm

SPEAKER’S BIO:

Giovanni Semeraro is Associate Professor at the University of Bari “Aldo
 Moro” since 1998, where he leads the research group SWAP (Semantic Web
 Access & Personalization, http://www.di.uniba.it/~swap/) at the Department
 of computer science. 
 His main research interests fall into the following areas:
 Intelligent Information Access, 
Recommender Systems
, Information Mining
, Machine Learning
, Personalization and User Modelling
, and Natural Language Processing
. He served as scientific responsible for 10 international and national
 projects and 16 research contracts. 
 He is editor of 8 international books and author of more than 300 scientific
 papers published in international journals, books, conference and workshop
 proceedings.

TALK ABSTRACT:

Content-based recommender systems (CBRS) analyze a set of objects, usually
 textual descriptions of items previously rated by a user, and build a model
 of user interests, called user profile, based on the features of the objects
 rated by that user. The user profile is then exploited to recommend new
 potentially relevant items.

In spite of the growing importance of collaborative filtering algorithms
 over the last years, Web 2.0 and the huge amount of user generated content,
 such as tags, annotations, folksonomies, etc., are providing new
 opportunities and challenges for CBRS.

The talk discusses the main problems which cause some limitations of CBRS,
 such as overspecialization and limited availability of content, and
 describes current research directions for overcoming them, including:

- defeating homophily in recommender systems: introducing serendipity for

recommendation diversification

- knowledge infusion into CBRS: exploiting open knowledge sources

(Wikipedia, folksonomies) for improving recommendation algorithms

- cross-language recommender systems: algorithms for learning multilingual

content-based profiles

REFERENCE PERSON: Francesco Ricci , Francesco.Ricci@inf.unibz.it


TIME: December 1, 2:30PM-4:30PM.

PLACE: Room F003, Main FUB Building, Sernesistrasse 1

SPEAKER: Martin Theobald (Max-Planck Institute for Informatics, Saarbruecken, Germany)

TALK TITLE:

Interactive Reasoning in Large and Uncertain RDF Knowledge Bases

SPEAKER’S HOMEPAGE:

http://www.mpi-inf.mpg.de/~mtb/

SPEAKER’S BIO:

Martin Theobald is a Senior Researcher at the Max-Planck Institute for Informatics. He obtained a doctoral degree in computer science from Saarland University in 2006 and spent two years as a post-doc at Stanford University, where he worked on the Trio probabilistic database system. Martin received an ACM SIGMOD dissertation award honorable mention in 2006 for his work on the TopX search engine for efficient ranked retrieval of semistructured XML data. His current research interests include information extraction, probabilistic databases and uncertain data management, as well as querying and ranking of semistructured data.

TALK ABSTRACT:

Recent advances in Web-based information extraction such as DBpedia and YAGO have devised the way for the automatic construction and growth of large, semantic knowledge bases. The very nature of these extraction techniques however entails that the resulting RDF knowledge bases may face a significant amount of incorrect, incomplete, or even inconsistent (i.e., uncertain) factual knowledge. This talk presents new results on extracting large-scale knowledge from free-text Web sources, and it introduces query-driven reasoning techniques that directly address the resolution of uncertainty within such data. Specifically, we present URDF, an efficient reasoning framework for uncertain RDF knowledge bases. URDF provides a SPARQL-like query model, and it combines rule-based, first-order predicate logic with probabilistic models to infer new facts and resolve data uncertainty. Moreover, our UViz frontend dynamically accesses and visually supports the URDF reasoning backend, thus providing an intuitive user interface for exploring the knowledge base, visualizing the steps involved in both the rule-based and probabilistic reasoning, as well as explaining answers through lineage.

REFERENCE PERSON: Mouna Kacimi , Mouna.Kacimi@unibz.it


TIME: November 29, 11:00AM-12:00PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Iván Cantador (Universidad Autónoma de Madrid, Madrid)

TALK TITLE:

Bringing Semantics to Folksonomies: Application to Item Recommendation

SPEAKER’S HOMEPAGE:

http://arantxa.ii.uam.es/~cantador/

SPEAKER’S BIO:

Iván Cantador is a lecturer at Universidad Autónoma de Madrid, where he obtained a PhD in computer science in 2008. During his doctoral studies, he was a research consultant at the Knowledge Media Institute of the Open University, and a research visitor at the University of Southampton. After earning his PhD, he worked for one year as a research assistant at the University of Glasgow. Dr. Cantador has participated actively in nine R&D projects, among which the aceMedia and MESH FP6 projects. His main research interests include Recommender Systems, Semantic Web, and Machine Learning. Recently, his research has focused on Information Retrieval and Social Systems, where he is investigating personalised and collaborative strategies for content retrieval and recommendation based on folksonomies. In the last five years, Iván has co-authored over forty publications in international journals and conferences such as SIGIR, ECIR, RecSys, ISWC, and Hypertext. He has been a reviewer in major journals such as Journal of Web Semantics, International Journal of Semantic Web and Information Systems, IEEE Transactions on Neural Networks, Computational Intelligence, and Computers & Mathematics with Applications, as well as a PC member in international conferences and workshops.

TALK ABSTRACT:

In the so called Web 2.0, systems facilitate the creation of diverse formats of user generated content. Among these content formats, social tagging has become a popular practice as a lightweight mean to classify and exchange information. Users create or upload content (items), annotate it with freely chosen words (tags), and share these annotations with others.

The whole set of annotations constitutes an unstructured collaborative knowledge classification scheme that is commonly known as folksonomy. This implicit classification serves various purposes, such as for item organization, promotions, sharing with friends, with the public, etc. Despite these advantages, tags are free text, and thus suffer from various vocabulary problems. Ambiguity of the tags arises as users apply the same tag in different domains and semantic contexts. At the opposite end, the lack of synonym control can lead to different tags being used for the same concept, precluding collocation. Moreover, multilinguality also obstructs the achievement of a consensus vocabulary, since several tags written in different languages can express the same concept.

To cope with the above problems, folksonomy-based information retrieval and filtering engines have to identify and exploit the underlying semantic meanings of the tags. In this talk, we present three approaches to bring semantics to folksonomies. The first approach automatically links social tags to semantic concepts existing in ontologies. Once the semantic concept of a tag has been identified, the second approach goes a step beyond and categorizes the tag based on the user’s tagging purposes. Finally, aiming to address the semantic ambiguity, the third approach selects the actual meaning of a tag within its particular annotation context. In the talk, we also explain how the above approaches have been evaluated and used to improve various item recommendation strategies.

REFERENCE PERSON: Francesco Ricci , Francesco.Ricci@inf.unibz.it


TIME: September 23, 3PM-4PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Shlomo Berkovsky (CSIRO, TasICT Centre, Australia)

TALK TITLE:

Group-Based Recipe Recommendations: Analysis of Data Aggregation Strategies

SPEAKER’S HOMEPAGE:

http://cs.haifa.ac.il/~slavax/

SPEAKER’S BIO:

Shlomo Berkovsky is a Research Team Leader at the Tailored Lifestyle Information project of the CSIRO Tasmanian ICT Centre. The project aims to provide individuals and groups of users with a personalized dietary and health information to help them to maintain a healthier lifestyle. Shlomo’s research interests include user modeling and personalization. In particular, he is interested in recommender systems, mediation of user models, ubiquitous user modeling, context-aware personalization, and personalized content generation. Shlomo received his PhD and MSc degrees from the University of Haifa and the topics of his theses were, respectively, “Mediation of user models for enhanced personalization in recommender systems” and “Unspecified ontologies for peer-to-peer e-commerce applications”. He is the author of more than 60 refereed publications accepted to journals, books, and conference proceedings.

TALK ABSTRACT:

Collaborative filtering recommendations were designed primarily for individual user models and recommendations. However, nowadays more and more scenarios evolve, in which the recommended items are consumed by groups of users rather than by individuals. This raises the need to uncover the most appropriate group-based collaborative filtering recommendation strategy. In this work we investigate the use of aggregated group data in collaborative filtering recipe recommendations. We present results of a study that exploits recipe ratings provided by families of users, in order to evaluate the accuracy of several group recommendation strategies and weighting models, and analyze the impact of switching strategies, data aggregation heuristics, and group characteristics on the performance of recommendations.

REFERENCE PERSON: Francesco Ricci , Francesco.Ricci@inf.unibz.it


TIME: July 1, 2PM-3PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Neil Rubens (University of Electro-Communications, Tokyo, Japan)

TALK TITLE:

Recommending Learning Actively

SPEAKER’S HOMEPAGE:

http://www.hrstc.org/

SPEAKER’S BIO:

Neil Rubens is an Assistant Professor at the Graduate School of Information Systems, University of Electro-Communications (Tokyo, Japan). He received Ph.D. in computer science from Tokyo Institute of Technology, and M.Sc. from the University of Massachusetts. He is also a visiting researcher and a member of Innovation Ecosystems Initiative at Media X, Stanford University. His research focuses on `Active Intelligence’. Unlike traditional Artificial Intelligence systems passively learn from the data that is fed to it; Active Intelligence systems actively seek and acquire additional relevant data throughout the learning process.

TALK ABSTRACT:

The problem in many relationships is that one side does not listen; this seems to be a problem in recommendation systems (RS) as well. Active learning can help the system to hear what the user wants. RSs tell the users what it thinks they want, but often do not actively engage in a process of finding out what that is. In other words, systems integrated with AL could provide the next level of personalization, the goal of RSs in the first place. However, we have still yet to see a wide adoption of active learning into recommender systems; one possible reason for this is that integration of the two is not necessarily straightforward. We explain what active learning is, how it can be used in recommender systems, and discuss the specific requirements and settings that must be considered when adapting it for such systems. Utilizing AL properly can create a system capable of creating an enjoyable experience of self-discovery and exploration for the user, while at the same time satisfying system objectives, like profitability, etc.

REFERENCE PERSON: Francesco Ricci , Francesco.Ricci@inf.unibz.it


Robin Burke

TIME: June15-17, 11AM-12PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Robin Burke (DePaul University)

TALK TITLE:

Social Computing

SPEAKER’S HOMEPAGE:

http://josquin.cti.depaul.edu/~rburke/

SPEAKER’S BIO:

Robin Burke is an associate professor at the School of Computing and Digital Media at DePaul University in Chicago, Illinois. He also held positions at the University of Chicago, the University of California, Irvine, and California State University, Fullerton after receiving his PhD from Northwestern University’s Institute for the Learning Sciences in 1993. Dr. Burke has been active in recommender systems research since 1995. His current research examines recommendation algorithms in the social web, and alternative evaluation methodologies for recommender systems.

TALK ABSTRACT:

Social computing is the application of computing to represent, model and support social interactions between people. Social computing applications such as Facebook and Twitter have become some of the most widely used computing applications ever devised. In this lecture series, we will explore three aspects of social computing systems.

* Informal knowledge engineering. This type of social computing is a feature of many sites, but stands out particularly in media sharing sites like flickr and Last.fm, and in dedicated tagging sites like delicious.

* Building, maintaining and using explicit connections between users. In sites like Facebook and LinkedIn, connections are used to distribute information, engage in shared activities (i.e. games), and maintain social ties.

*Organizing exchanges in which individuals compete to maximize their utility. Of course, the world’s stock and commodities exchanges have long been computerized, but more recently, open electronic markets have gained wide acceptance in Google’s AdWords and the auction site eBay.

Each of these areas has seen major practical and theoretical advances over the last decade. Computing has become, more than ever before, a vehicle for individual expression and for the maintenance of social ties. We will examine this phenomenon from a computer science perspective, examining specific results and systems, general research trends, and open problems.

REFERENCE PERSON: Francesco Ricci , Francesco.Ricci@inf.unibz.it


Florian Michahelles

TIME: Tuesday, May 4, 9:30AM-10:30A3PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Florian Michahelles (ETH Zurich)

TALK TITLE:

Towards a More Tangible Search of Information: Internet of Things

SPEAKER’S HOMEPAGE:

http://www.im.ethz.ch/people/fmichahelles

 

 

SPEAKER’S BIO:

Florian Michahelles is the associate director of the Auto-ID Labs Zurich/St. Gallen and manager of the Information Management lab of Prof. Fleisch at ETH Zurich. His research interests comprise the conception, development and the evaluation of the electronic integration of daily-life objects and the connected, independent exchange of information between these objects (internet of things) for the private end-user as well as for commercial use in economic surroundings. Michahelles received a PhD from ETH Zurich for his research in participative design of wearable computing applications and the development of innovative business cases for ubiquitous computing. He holds a M.Sc (Diplom-Informatiker Univ.) degree in computer science and psychology from the Ludwig-Maximilians-University of Munich. Michahelles has published 50+ papers in international journals (e.g. IEEE Pervasive Computing, computer & Graphics), conferences and scientific workshops.

TALK ABSTRACT:

In the vision of the Internet of Things physical items of the world become empowered by computing (smart things), or become at least identifiable (tagging) for tracking and for aligning information to single units. This emerging networked reality providing computers with senses about what is going on in the real-world aims at providing a much more error-free transfer of information: whenever people have to report on phenomena, manually enter data, or even operate support systems as barcode scanners, they introduce errors to their input data. Instead, the internet of things offers both the opportunity to let things themselves report about their status and location as well to embed information seamlessly into our world, attach it to objects, places, people etc., rather than keeping information separate from our world in isolated computers. This gives access to information directly at the phenomena where needed.

This talk will illustrate various approaches about how information can be accessed in a more intuitive way than traditional computing provides, .e.g.: instead of searching for information by user defined search parameters, sensors in the user’s mobile phone can derive the user’s current location and feed that data into a query to determine relevant information. This talk will provide an overview of various research projects going on at ETH Zurich and discuss new opportunities and emerging challenges.

REFERENCE PERSON: Francesco Ricci , Francesco.Ricci@inf.unibz.it


Floriano Zini

TIME: Thursday, April 1, 2PM-3PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Floriano Zini (University of Bolzano)

TALK TITLE:

Evaluation of Scheduling and Replica Optimization Strategies for a Data Grid

SPEAKER’S HOMEPAGE:

http://sra.itc.it/people/zini/

 

 

SPEAKER’S BIO:

Floriano Zini has been involved in IT R&D for more than 13 year, both in academia and industry. He is currently an assistant professor at the faculty of computer science of the Free University of Bozen – Bolzano, working in the Database and Information Systems research group. Zini holds a Ph.D. in computer science and has international research experience and has co-authored more than 40 papers and technical reports in machine learning, agent oriented software engineering and grid computing. He has also acquired practical experience in information retrieval and natural language processing by working for Expert System, an Italian leader company in these fields.

TALK ABSTRACT:

Data Grids have emerged in the last decade as a viable solution to the large computational power and data storage requirements of many e-science projects such as the next generation of high energy physics experiments recently started at CERN. Data Grids enable sharing of distributed computational and storage resources among users located all over the world.

An important open research issue is how to optimize the use of available Data Grid resources so that user jobs, which typically need to analyze huge amount of data, can be rapidly executed without being too resource-comsuming. Efficient job scheduling, i.e. the decision of when and where to run jobs submitted to a Data Grid, is important to ensure that resources are neither over- nor under-used. Data Replication – the process of creating identical copies of data files at different sites – is also an important part of maximisng job throughput in a typical Data Grid.

In this talk I present a research, performed in the framework of the European DataGrid project, that studies the effects of various job scheduling and data replication strategies. Such strategies are compared in a variety of Grid scenarios using several performance metrics. We use the Grid simulator OptorSim, and base our simulations on a world-wide Grid testbed for data intensive high energy physics experiments.

Our results show that scheduling algorithms which take into account both the file access cost of jobs and the workload of computing resources are the most effective at optimizing computing and storage resources as well as improving the job throughput. The results also show that, in most cases, the economy-based replication strategies which we have developed improve the Grid performance under changing network loads.

REFERENCE PERSON: Floriano Zini , floriano.zini@inf.unibz.it


Juozas Gordevicius

TIME: Thursday, March 18, 2PM-3PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Juozas Gordevicius (University of Bolzano)

TALK TITLE:

Ranking of Evolving Stories Through Meta-Aggregation

SPEAKER’S HOMEPAGE:

http://www.inf.unibz.it/~gordevicius/

 

 

SPEAKER’S BIO:

Juozas Gordevicius graduated, in 2004, with BSc in computer science degree from Vilnius University. Since then he continues his studies at Free University of Bozen Bolzano under supervision of prof. Johann Gamper. In 2006 he defended his MSc thesis and enrolled into PhD program. His main interest is issues in temporal databases, in particular temporal aggregation. In 2009 he collaborated with Thoora Inc. in Canada where he studied ranking problems in temporal data. The talk will be about the outcome of this collaboration.

TALK ABSTRACT: Since the beginning of the world wide web, interlinked sources of information have been the subject of active research in the knowledge discovery community. Several approaches have been proposed to deal with the challenging problems of information summarization and ranking using link analysis. In this paper, we leverage these previous approaches and focus on the problem of ranking news stories within their historical context by exploiting their content similarity. We treat news stories as evolving events over time and rank them in a time and query dependent manner. We do this in two steps. First, the mining step discovers metastories — meaningful groups of similar stories that occur at arbitrary points in time. Second, the ranking step uses well known measures of content similarity to construct implicit links among all metastories, and uses these to produce a ranking that is specific to the time interval specified in the user query. We use real data from both conventional and social media sources (weblogs) to study the impact of different meta-aggregation techniques, as well as similarity measures in the final ranking. These results are evaluated using both subjective and objective measures, and provide a useful guideline for the generation and ranking of metastories.

REFERENCE PERSON: Juozas Gordevicius , gordevicius@inf.unibz.it


Augsten Nikolaus

TIME: Wednesday, February 17, 3PM-4PM.

PLACE: Room 1.02, Dominikanerplatz 3 – Piazza Domenicani, 3

SPEAKER: Dr. Augsten Nikolaus (University of Bolzano)

TALK TITLE:

TASM: Top-k Approximate Subtree Matching

SPEAKER’S HOMEPAGE:

http://www.inf.unibz.it/~augsten/

 

 

SPEAKER’S BIO:

Nikolaus Augsten is an assistant professor in computer science at the Free University of Bozen-Bolzano. He received his PhD from Aalborg University, Denmark, in 2008. His research interests include data-centric applications in database and information systems with a particular focus on approximate matching techniques for complex data structures, efficient index structures for distance computations, and similarity search in massive data collections.

TALK ABSTRACT: We consider the Top-k Approximate Subtree Matching (TASM) problem: finding the k best matches of a small query tree, e.g., a DBLP article with 15 nodes, in a large document tree, e.g., DBLP with 26M nodes, using the canonical tree edit distance as a similarity measure between subtrees. Evaluating the tree edit distance for large XML trees is difficult: the best known algorithms have cubic runtime and quadratic space complexity, and, thus, do not scale. Our solution is TASM-postorder, a memory-efficient and scalable TASM algorithm. We prove an upper-bound for the maximum subtree size for which the tree edit distance needs to be evaluated. The upper bound depends on the query and is independent of the document size and structure. A core problem is to efficiently prune subtrees that are above this size threshold. We develop an algorithm based on the prefix ring buffer that allows us to prune all subtrees above the threshold in a single postorder scan of the document. The size of the prefix ring buffer is linear in the threshold. As a result, the space complexity of TASM-postorder depends only on k and the query size, and the runtime of TASM-postorder is linear in the size of the document. Our experimental evaluation on large synthetic and real XML documents confirms our analytic results. This work received the ICDE 2010 Best Paper Award

REFERENCE PERSON: Dr. Augsten Nikolaus, augsten@inf.unibz.it


Bhaskar Mehta

TIME: Friday, December 11, 11AM-12AM

PLACE: Room A101, Universitätsplatz 1 – piazza Università, 1, Bozen-Bolzano

SPEAKER: Dr. Bhaskar Mehta (Google Switzerland GmbH)

TALK TITLE:

Recommendations at Google

SPEAKER’S BIO:

Bhaskar Mehta is a computer scientist and software engineer at Google Zurich. His research has focused on recommender systems, and various ways to improve their accuracy, robustness and unbaisedness. He holds a PhD in CS (Uni Duisburg, with Thomas Hofmann and Norbert Fuhr) and a B.Tech and M.Tech degree in CSE (IIT Delhi, thesis with S.N. Maheshwari). His general research interests are Personalization and Recommender systems, Web Search algorithms and Web models, P2P and grid systems; data mining and statistical learning in application to Social media, social networks and personalized ranking. Book:Cross System Personalization

SPEAKER’S HOMEPAGE:

http://research.google.com/pubs/author36288.html

TALK ABSTRACT:

Google has been providing recommendations to millions of user, whether its on Youtube, Google Video, Product Search, Google Reader or implicitly on search. The challenges of scale and data sparsity are unique to our setting, and in this talk, I’ll focus on some of the major challenges and how we deal with them. The talk will cover basic data mining algorithms which are relevant to the problem, and some insight into potential research directions for the academic community.

REFERENCE PERSON: Prof. Francesco Ricci, Francesco.Ricci@unibz.it


Sven Helmer

TIME: Wednesday, December 9, 1PM-3PM

PLACE: Room C3.06, Universitätsplatz 1 – piazza Università, 1, Bozen-Bolzano

SPEAKER: Sven Hemler(University of London)

TALK TITLE:

XML Message Management

SPEAKER’S BIO:

Sven Helmer is currently a Senior Lecturer at the Department of computer science and Information Systems at Birkbeck, University of London. He obtained a PhD from the University of Mannheim, Germany, and an MSc in computer science from the University of Karlsruhe, Germany. He has also held a visiting professorship in databases at the University of Heidelberg, Germany. His research interests include native XML databases, query optimization, multi-user synchronization as well as interdisciplinary research in the areas of astronomy, physics, and ethnography.

He has published more than 40 peer-reviewed papers and book chapters.

SPEAKER’S HOMEPAGE:

http://www.dcs.bbk.ac.uk/~sven/

TALK ABSTRACT:

The eXtensible Markup Language (XML) has spread rapidly as a language for exchanging information in many different application areas such as financial services, biology, and news services. We present a new model, based on message queues, to process XML data natively in the context of a service-oriented architecture (SOA).

After discussing the downsides of the current approaches, we highlight some of the features of this new model and identify the challenges in turning this approach into a highly scalable, distributed system.

REFERENCE PERSON: Prof. Johann Gamper , gamper@inf.unibz.it


Peer Kröger

TIME: Friday, December 2, 09:30AM-10:30AM

PLACE: Room D002, Universitätsplatz 1 – piazza Università, 1, Bozen-Bolzano

SPEAKER: Peer Kröger (Ludwig-Maximilians-Universität München)

TALK TITLE:

How to find clusters in high-dimensional spaces: problems, solutions, and perspectives

SPEAKER’S BIO:

Peer Kröger has a tenured position as Akademischer Rat in the database systems and data mining group at the Ludwig-Maximilians-Universitaet Muenchen, Germany. He finished his PhD thesis on clustering moderate-to-high dimensional data in July 2004 and his Habilitation on data mining and similarity search in scientific data in January 2009. He published more than 60 papers in peer reviewed conferences and journals and received the “Best Paper Honorable Mention Award” from the SIAM International Conference on Data Mining 2008. In addition, he is a co-author of several tutorials and

served as program committee member as well as reviewer of the major database and data mining conferences and journals. His research interests include mining and searching multimedia databases.

SPEAKER’S HOMEPAGE:

http://www.dbs.informatik.uni-muenchen.de/cms/Peer_Kr%C3%B6ger

TALK ABSTRACT:

Modern capabilities of generating and recording information leads to a

huge amount of high-dimensional data stored in massive databases. Data

mining methods provide the potentials to make use of all this data by

automatically deriving previous unknown knowledge.

High-dimensional data, however, poses some special challenges for data

mining methods that require specialized solutions. In this talk, basic

problems of high dimensional data, often summarized by the term “curse

of dimensionality”, are shortly reviewed and general solutions are

sketched. It turns out that clusters are more likely to be detected in

subspace of the original feature space. An algorithm for this problem is

presented that generalizes the concept of the Hough transform for

finding linear structures in images. Finally, properties of this

algorithm are discussed and potential directions for future work are shown.

REFERENCE PERSON: Prof. Johann Gamper , gamper@inf.unibz.it


Alejandro Vaisman

TIME: Tuesday, December 1, 2PM-3PM

PLACE: Room D003, Universitätsplatz 1 – piazza Università, 1, Bozen-Bolzano

SPEAKER: Alejandro Vaisman (University of Buenos Aires-FCEN, Argentina)

TALK TITLE:

Integrating Spatial, OLAP, and Moving Object Data

SPEAKER’S BIO:

Alejandro Vaisman was born in Buenos Aires (UBA), Argentina. He received a BA degree in Civil Engineering, a BA in Computer Science, and a PhD in Computer Science from UBA, and he has been a post-doctoral researcher at the University of Toronto. He is a Professor at the UBA since 1994. He

was an invited professor at the Universidad Politecnica de Madrid in 1997. In 2001 he was appointed Vice-Dean of the School of Engineering and Information Technology at the University of Belgrano, in Argentina. He was a visiting researcher at the University of Toronto, University of Hasselt and Universidad de Chile. His research interests are in the field of databases, particularly in OLAP, Data Mining, XML and the Semantic Web, and Geographic Information Systems. In 2004 he was appointed Vice-Head of the Department of Computer Science, and Chair of the Graduate Program in Data Mining at the Computer Science Department, UBA. He is the main responsible of the project “Using OLAP Techniques in Geographical Information Systems”, funded by the Argentinian Scientific Agency.

SPEAKER’S HOMEPAGE:

http://www.cs.toronto.edu/~avaisman/

TALK ABSTRACT:

In this talk I will present an approach to solve the problem of integrating spatial data (stored in a GIS),

OLAP data (stored in a data warehouse), and moving object data. For this, I will first introduce a formal data model, denoted Piet, that integrates GIS and OLAP in a unique framework. Piet is equipped with an SQL-like query language denoted Piet-QL, that supports four basic kinds of queries: (a) standard spatial queries; (b) standard OLAP queries; (c) spatial queries filtered with an aggregation (i.e., filtered using a data cube); (d) OLAP queries filtered with a spatial condition. The integration mentioned above becomes more interesting when moving objects are included in the picture. The notion of semantic trajectories (i.e., trajectories expressed in terms of places of interest instead of x,y coordinates ) allows inferring interesting patterns of movement. Usually, this pattern analysis is performed disregarding the information about the geographic location where the moving objects evolve. I will present a language that can intensionally express sequential patterns by means of regular expressions built over constraints defined over the attributes of the places of interest visited by the trajectories under analysis. I will also present our implementation of theaforementioned model and languages. Finally, I will comment on current and future research directions.

REFERENCE PERSON: Prof. Johann Gamper , gamper@inf.unibz.it


Alexander Felfernig

TIME: Friday, November 27, 9AM-10AM

PLACE: Room F001, Universitätsplatz 1 – piazza Università, 1, Bozen-Bolzano

SPEAKER: Prof. DI Dr. Alexander Felfernig(Graz University of Technology)

TALK TITLE:

Effective Knowledge Engineering for Constraint-based Systems

SPEAKER’S BIO:

Alexander Felfernig is university professor at the Graz University of Technology (TU Graz). He is the head of the research group “Applied Software Engineering” (ASE) that focuses on research in the areas of intelligent user interfaces for constraint-based applications, testing & debugging methods for knowledge bases, modeling approaches for complex knowledge bases, and related decision and cognitive psychological aspects. This research has been honored this year with the “Heinz Zemanek Award” of the Austrian Computer Society. Alexander Felfernig has published over 120 research papers in the mentioned research fields, acts as co-organizer of international scientific workshops and conferences, and is editor of related journal issues. He is a co-founder and managing director of ConfigWorks, a company that focuses on the development of knowledge-based configurator and recommender applications.

SPEAKER’S HOMEPAGE:

http://www.felfernig.eu

TALK ABSTRACT:

“Constraint programming represents one of the closest approaches computer science has yet made to the Holy Grail of programming: the user states the problem, the computer solves it” (E. Freuder, Constraints Journal, 2: 57-61, 1997).

Nowadays constraint technologies are successfully applied in many fields such as configuration, scheduling, recommendation, and resource allocation. However, knowledge engineers are often overwhelmed by the size and complexity as well as the increasing change rates of constraint knowledge bases. This situation creates enormous challenges on an in-time provision of up-to-date and quality-assured applications. The talk will focus on intelligent testing and debugging techniques which support knowledge engineers in the development & maintenance of elementary parts of constraint-based applications: user interface descriptions, rule sets for the estimation of user preferences, and core knowledge bases. In this context, knowledge-based configurators will serve as a working example. The talk will be concluded with an outlook on new application domains and research challenges in the field.

REFERENCE PERSON: Prof. Francesco Ricci, Francesco.Ricci@unibz.it


Sherif Sakr

TIME: Monday, November 16, 11AM-12AM

PLACE: Room A101, Universitätsplatz 1 – piazza Università, 1, Bozen-Bolzano

SPEAKER: Dr. Sherif Sakr (University of New South Wales (UNSW), Australia)

TALK TITLE:

GraphREL: A relational approach for scalable processing of subgraph queries

SPEAKER’S BIO:

Dr. Sherif Sakr is a senior research associate / lecturer in the Service Oriented Computing (SOC) research group at School of Computer Science and Engineering (CSE), University of New South Wales (UNSW), Australia. He received his PhD degree in computer science from Konstanz University, Germany in 2007. He received his BSc and MSc degree in computer science from information systems department, the Faculty of Computers and Information, Cairo University, Egypt, in 2000 and 2003 respectively. Prior to taking up the current position, he worked as a postdoctoral research fellow at National ICT Australia (NICTA).

Much of Sherif’s past research work has revolved around the Pathfinder XQuery compiler which is cited by many researchers around the world.

Pathfinder compiles expressions from the W3C XQuery language into a text bookstyle relational algebra, enabling traditional relational database back-ends to operate as high-performance XQuery processors. Pathfinder is distributed as MonetDB/XQuery, together with its primary database back-end MonetDB. In addition, Pathfinder provides support for SQL:1999-compliant databases, as well as for the commercial main-memory database kdb+.

Sherif’s current research interests lie in the areas of Graph Data Management, Data Centers, Data Management in Cloud Computing and Business Process Modeling and Management. His work has been published in international journals and conferences such as: Proceedings of the VLDB endowment (PVLDB), Journal of Computer, Systems and Science (JCSS), Journal of Database Management (JDM), International Journal of Web Information Systems (IJWIS), VLDB, SIGMOD, DASFAA, and DEXA. One of his papers has awarded the Outstanding Paper Excellence Award 2009 of Emerald Literati Network. Dr. Sakr is also a reviewer for IEEE TKDE, IEEE TSC, DKE and IJWET journals and many international conferences.

SPEAKER’S HOMEPAGE:

http://www.cse.unsw.edu.au/~ssakr/

TALK ABSTRACT:

Graphs are widely used for modelling complicated data such as: chemical compounds, protein interactions, social networks and semantic web.

Retrieving related graphs containing a query graph from a large graph database is a key issue in any graph-based applications. Relational database management systems (RDBMSs) have repeatedly been shown to be able to efficiently host different types of data which were not formerly anticipated to reside within relational databases such as complex objects and XML data. RDMBSs derive much of their performance from sophisticated optimizer components which makes use of physical properties that are specific to the relational model such as: sortedness, proper join ordering and powerful indexing mechanisms. In this talk, we give an overview of the problem of indexing and querying graph databases. We present a novel, decomposition-based and selectivity-aware approach for indexing and querying graph database using the relational infrastructure. The proposed approach carefully exploit existing database functionality such as partitioned B-trees indexes and influencing the relational query optimizers by selectivity annotations to reduce the access costs of the secondary storage to a minimum. A discussion of how this problem can be generalized to other types of graphs and how it can utilize from the recently introduced Map/Reduce framework will be also included.

REFERENCE PERSON: Prof. Johann Gamper , gamper@inf.unibz.it


Markus Zanker

TIME: Tuesday, November 10, 9AM-10AM

PLACE: Room A101, Universitätsplatz 1 – piazza Università, 1, Bozen-Bolzano

SPEAKER: Prof. Markus Zanker (Department for Applied Informatics, University of Klagenfurt, Austria)

TALK TITLE:

Harnessing Geotagged Resources for Web Personalization

SPEAKER’S BIO:

Markus Zanker is an assistant professor in the Department for

Applied Informatics and director of the study program Information

Management at the University of Klagenfurt. His research interests focus

on knowledge-based systems, particularly in the fields of interactive

sales applications, such as product configuration and recommendation. He

also works on knowledge acquisition and user modeling for

personalization. In order to apply research results to industry he has

co-founded ConfigWorks GmbH, a provider of interactive selling

solutions.

SPEAKER’S HOMEPAGE:

http://www.configworks.com/mz/index.html

TALK ABSTRACT:

Due to the high availability of mobile devices with GPS functionality

among consumers, social web platforms cumulate more and more geotagged

webresources. This talk therefore presents an approach that harnesses

these resources by automatically deriving semantic knowledge from

geocoded resources and exploiting them for personalization. The utility

of the approach is demonstrated by an adaptive WebGIS scenario.

REFERENCE PERSON: Prof. Francesco Ricci, Francesco.Ricci@unibz.it


Lior Rokach

TIME: Tuesday, November 3, 9AM-10AM

PLACE: Room A101, Universitätsplatz 1 – piazza Università, 1, Bozen-Bolzano

SPEAKER: Prof. Lior Rokach(Department of Information Systems Engineering Ben-Gurion University of the Negev Beer-Sheva, Israel)

TALK TITLE:

Active Learning for Preferences Elicitation in Recommender Systems

SPEAKER’S BIO:

Dr. Lior Rokach is a senior lecturer at the Department of Information System Engineering at Ben Gurion University and the software engineering program. Dr. Rokach is a recognized expert in intelligent information systems and has held several leading positions in the industry directly related to this field. His main areas of interest are data mining, pattern recognition, and information retrieval.

Dr. Rokach is the author of over 70 refereed papers in leading journals (e.g. Data Mining and Knowledge Discovery, IEEE Transactions on Knowledge and Data Engineering and Pattern Recognition), conference proceedings and book chapters. In addition, he has also authored six books including Pattern Classification Using Ensemble Methods (World Scientific Publishing, 2009), Data Mining with Decision Trees (World Scientific Publishing, 2007) and Decomposition Methodology for Knowledge Discovery and Data Mining (World Scientific Publishing, 2005).

Dr. Rokach is also the co-editor of The Data Mining and Knowledge Discovery Handbook (Springer, 2005), Soft Computing for Knowledge Discovery and Data Mining (Springer, 2007) and Recommender Systems Handbook (Springer, 2010).

Dr. Rokach holds a B.Sc., M.Sc. and PhD in Industrial Engineering from Tel Aviv University.

SPEAKER’S HOMEPAGE:

http://www.ise.bgu.ac.il/faculty/liorr/

TALK ABSTRACT:

Active learning techniques construct accurate prediction models while minimizing the acquisition cost of labels. They achieve this by actively selecting cases that obtain the best improvement of the model. In this research, we utilize active learning for the construction of recommender models. We present a new probabilistic-based method for acquiring the user preferences in order to improve the predictive performance of the recommender system. Based on the user’s implicit response, the learner selects the next items to be presented to the user, and so forth. The new method can also be used to balance the exploration/exploitation trade-off. A large and prolonged field study was conducted with the cooperation of a well-established communication enterprise.

The proposed method achieves objective and subjective results that are both superior to current state-of-the-art methods. This renders the method suitable to alleviate the cold-start problem.

REFERENCE PERSON: Prof. Francesco Ricci, Francesco.Ricci@unibz.it


Boris Glavic

TIME: Wednesday, October 21, 9AM-10AM

PLACE: Room A101, Universitätsplatz 1 – piazza Università, 1, Bozen-Bolzano

SPEAKER: Boris Glavic (University Zurich)

TALK TITLE:

Integrating Data Provenance Support in Database Systems

SPEAKER’S BIO:

Boris Glavic received a diploma in Computer Science from theRWTH Aachen (Germany) in 2005. Since then he is a Ph.D. student at the University Zurich’s Database Technology Research Group. His main research interest is Data Provenance with a focus on integrating provenance support in database systems. This work is a collaboration with Gustavo Alonso (ETH Zurich) and has been published

at the EDBT and ICDE.

SPEAKER’S HOMEPAGE:

http://www.ifi.uzh.ch/dbtg/Staff/Glavic/

TALK ABSTRACT:

Data provenance, information about the origin and creation process of data, plays an important role in E-Science, Datawarehousing, Workflow-Management systems, and Grid computing. The concept of provenance in the context of relational database has been studied extensively from a theoretical point of view, but there is a apparent lack of ‘real’ provenance management systems. Such a system should be able to store and generate provenance information efficiently and expose this information to the user via an expressive query language. In this talk I present Perm, a provenance management system developed at University Zurich, that uses query rewrite techniques to generate provenance information. The talk will highlight the formal aspects of the approach as well as demonstrating how Perm was implemented as an extension of PostgreSQL.

REFERENCE PERSON: Prof. Johann Gamper , gamper@inf.unibz.it


Robin Burke

TIME: Friday, June 26, 11AM-12PM

PLACE: A101 (Sernesi building)

SPEAKER: Prof. Robin Burke (DePaul University, Chicago, Illinois, USA)

TALK TITLE:

Robust Recommender Systems

SPEAKER’S BIO:

Robin Burke is a 2008-2009 Fulbright Scholar at University College

Dublin and an associate professor at the School of Computing and Digital

Media at DePaul University in Chicago, Illinois. He also held positions

at the University of Chicago, the University of California, Irvine, and

California State University, Fullerton after receiving his PhD from

Northwestern University’s Institute for the Learning Sciences in 1993.

Dr. Burke has been active in recommender systems research since 1995,

and was a pioneer in the area of case-based approaches to

knowledge-based recommender systems. His current research examines

security properties of recommendation algorithms and alternative

evaluation methodologies for recommender systems.

SPEAKER’S HOMEPAGE:

http://gamejam.cti.depaul.edu/~rburke/

TALK ABSTRACT:

The openness and anonymity of the Internet environment create many

hazards for e-commerce. For collaborative recommender systems, it raises

the possibility of that attackers will seek to bias the output

recommendations through manipulation of the public inputs that the

system permits. Fighting such manipulation is a constant battle for the

owners and maintainers of such systems. In this talk, I will describe

the known vulnerabilities of collaborative algorithms and examine a

range of possible attack types that could be deployed against them. With

these vulnerabilities in mind, I will discuss possible responses,

including the deployment of alternate recommendation algorithms and the

use of supervised and unsupervised techniques to detect attacks.

Building on this research, I will examine what it might mean to build a

robust collaborative recommender and consider the implications for other

machine learning techniques deployed in public on-line environments.

REFERENCE PERSON: Prof. Francesco Ricci, Francesco.Ricci@unibz.it


TIME: Monday, June 22, 9AM-10AM

PLACE: A101 (Sernesi building)

SPEAKER: Dr. Federica Paci (University of Trento, Italy)

TALK TITLE:

An Integrated Digital Identity and Access Management Solution for Business Processes

SPEAKER’S BIO:

Dr. Paci is now a post-doc at Department of Engineering and Computer Science of the University of Trento. She works in the Secure Change European project. From February 2008 to March 2009 Dr. Federica Paci was a post-doctoral research associate at Purdue University. Paci’s main research interests include access control for service oriented architectures and for virtual organizations and trust negotiations. Currently, she is exploring security issues in the context of social networks and is developing trust negotiation protocols in peer-to-peer platforms. Paci earned her Ph.D. in Computer Science from the University of Milan, Italy, in February 2008. In February 2004, she received the equivalent of a combined bachelor’s/master’s degree in Computer Science, also from the University of Milan. During Spring Semester of 2005 and 2006, Paci was a visiting research scholar at CS Department of Purdue University, West Lafayette, IN. Paci is the author or co-author of more than 15 conference papers and journal articles. Currently, she is co-authoring a book on Web services security.

She serves as a program committee member for APWeb 2008 , IEEE Collaborative-Com 2008, WWW 2009.

SPEAKER’S HOMEPAGE:

http://www.dit.unitn.it/users/federica.paci

TALK ABSTRACT:

Business processes have attracted considerable research interest in the last fifteen years. More recently, several XML-based languages have been proposed for specifying and orchestrating business processes, resulting in the WS-BPEL language. WS-BPEL has been developed to specify automated business processes that orchestrate activities of multiple Web services. There are, however, cases in which people must be considered as additional participants to the execution of a process. Therefore, it is important to extend WS-BPEL to include the specification of activities that must be fully or partially performed by humans. The inclusion of humans, in turn, requires solutions for verifying the identity of users who request the execution of human activities and for the specification and enforcement of authorizations to users for the execution of human activities while enforcing constraints, such as separation of duty, on the execution of those activities. Another important requirement related to the inclusion of humans is resiliency to user unavailability. The set of available users may change during the execution of a WS-BPEL process for a large variety of reasons, and therefore, it is important to verify a process can complete even if certain users become unavailable.

This talk presents an integrated digital identity and access control management solution for WS-BPEL processes. We present RBAC-WS-BPEL, an RBAC model for WS-BPEL business processes, which supports a privacy-preserving verification of users identity attributes, the specification and enforcement of resiliency constraints, authorizations and authorization constraints on business process human activities.

Resiliency constraints are evaluated when a WS-BPEL process is deployed, to check if there are a sufficient number of authorized users to perform the process so that authorization constraints are satisfied and the process terminates even if some users become unavailable.

The identity of a user is verified when he/she requests the execution of an activity in order to determine whether the user has the permission to perform the activity. The privacy-preserving identity verification process allows a user to prove the knowledge of multiple identity attributes stored as cryptographic commitments using aggregated zero knowledge proof of knowledge (ZKPK) and Oblivious Commitment-Based Envelope (OCBE) protocols.

Once the user’s identity has been verified, authorizations and authorization constraints are evaluated to verify that the execution of the activity by the user does not violate any authorization constraints and does not prevent some other subsequent activities from completing.

The talk, in addition to presenting definitions and algorithms, will present the architecture of a system implementing the proposed solution and will report experimental results about the system performance.

REFERENCE PERSON: Prof. Johann Gamper, gamper@inf.unibz.it


Bjørn Zenker

TIME: Wednesday, June 3, 3PM-4PM

PLACE: A 101 (Sernesi building)

SPEAKER: Dipl. Inf. Bjørn Zenker (University of Erlangen-Nürnberg, Germany)

TALK TITLE:

ROSE

SPEAKER’S BIO:

Bjørn Zenker graduated in computer science 2008 at the Friedrich-Alexander-University Erlangen-Nürnberg. In his thesis he developed the decision-based programming language MADL for multi-attributive decision making. Since october 2008 he is working an assistant professor for the Artificial Intelligence Division of the University of Erlangen-Nürnberg in the ROSE team. His main research interest is the coupling of recommender technology with multi destination route generation by considering user preferences and decision strategies.

SPEAKER’S HOMEPAGE:

http://www8.informatik.uni-erlangen.de/en/zenker.html

TALK ABSTRACT:

ROSE (ROuting SErvice) is a mobile recommendation and navigation system for pedestrians with public transport support. Our motivation is to free the passenger from as many tedious tasks as possible. The user does not even have to know the destination! This is because ROSE will automatically propose events and destinations matching the interests of the user. For example, the user could ask if there are any outdoor music events today and ROSE will search huge databases, provided by our partners for available events and propose them to the user. Matching is done by computing a multi-attribute optimization that compares features of a proposal with the user model.

It also determines the best possible transport link and accompanies passengers throughout their entire journey by taking live public transport data into account.

REFERENCE PERSON: Prof. Francesco Ricci, Francesco.Ricci@unibz.it


Gerasimos Marketos

TIME: Friday, March 20, 3PM-4PM

PLACE: C 2.6 (Sernesi building)

SPEAKER: Gerasimos Marketos (University of Piraeus, Greece)

TALK TITLE:

Analyzing Trajectory Data

SPEAKER’S BIO:

I am a PhD candidate at the Department of Informatics, University of Piraeus

(UniPi), Greece. I received my Bachelor of Science Degree (2003) in

Informatics from University of Piraeus and my Master of Science Degree

(2004) in Information Systems Engineering from University of Manchester

Institute of Science and Technology (UMIST), UK. My research interests

include spatiotemporal data warehousing and mining, pattern management and

scientific databases. Currently, I participate in the EC-funded GeoPKDD

project (2005-09) on geographic privacy-aware knowledge discovery and

delivery and I am also involved in several national-level projects.

URL: http://infolab.cs.unipi.gr/people/marketos/

SPEAKER’S HOMEPAGE:

http://infolab.cs.unipi.gr/people/marketos/

TALK ABSTRACT:

This talk will discuss techniques for extending Data Warehousing and Data

Mining technologies so as to apply them on moving object data

(trajectories). The analysis of such trajectory data raises opportunities

for discovering behavioral patterns that can be exploited in applications

like service accessibility, mobile marketing and traffic management.

Trajectory Data Warehousing (TDW) aims at transforming raw trajectory data

to knowledge about moving objects and at providing summarized trajectory

information. Modeling, ETL (trajectory reconstruction, cube loading) and

OLAP (aggregation) issues will be discussed.

Traffic Mining (TM) is a special kind of trajectory mining area as it is

focused on discovering knowledge that can be used for better controlling

traffic and optimizing traffic flows. We will focus on the utilization of

graph techniques in order to analyze the traffic flow in the network and on

discovering traffic relationships like propagation, split and merge of

traffic among the road segments.

REFERENCE PERSON: Prof. Johann Gamper, gamper@inf.unibz.it


Giro Cattuto

TIME: Tuesday, February 17, 11:00AM-12:00PM

PLACE: A101 (Sernesi Building)

SPEAKER: Prof. Ciro Cattuto (Institute for Scientific Interchange (ISI Foundation), Torino)

TALK TITLE:

Measuring and Grounding Similarity from Social Annotations

SPEAKER’S BIO:

Ciro Cattuto is a Research Scientist at the Institute for Scientific Interchange (ISI Foundation) in Torino. He received his PhD in Physics from the University of Perugia (Italy). He subsequently worked at the University of Michigan in Ann Arbor (USA) and at the Frontier Research System of the RIKEN Institute (Japan) as a Fellow of the Japan Society for the Promotion of Science. After that he moved to the Sapienza University of Roma with a grant from the Enrico Fermi Center, where he played a key role in the creation of the TAGora EU project. In 2008 he moved to the Complex Network Lagrange Laboratory of the ISI Foundation. His current research focuses on complex phenomena in online information systems, and in general on using concepts of statistical physics and complex networks theory to study activity patterns and emergent properties in technological and social systems.

SPEAKER’S HOMEPAGE:

http://isiosf.isi.it/~cattuto

TALK ABSTRACT:

Social annotations are becoming increasingly important data sources for a variety of applications such as user profiling, community detection, navigation support, semantic search, and ontology learning.

A key problem is how to extend and adapt traditional notions of similarity to large bodies of lightweight and noisy annotations from social media.

In the first part of the talk we discuss a systematic characterization and validation of tag similarity in terms of formal representations of knowledge.

Using data from the social bookmarking system delicious.com, we investigate several measures of tag similarity. We provide a semantic grounding by mapping pairs of similar tags in the folksonomy to pairs of synsets in Wordnet, where we use validated measures of semantic distance to characterize the relation between the mapped terms. This exposes important features of the investigated notions of tag similarity, and indicates which of them are better suited in the context of a given application.

In the second part of the talk we discuss an evaluation framework to compare various folksonomy-based similarity measures derived from established information-theoretic, statistical, and practical measures.

We deal symmetrically with users, tags, and resources, and consider different methods to aggregate annotations across users.

We provide an external grounding based on WordNet and on the Open Directory Project, and investigate the issue of scalability.

REFS:

* “Semantic Grounding of Tag Relatedness in Social Bookmarking Systems”

C. Cattuto, D. Benz, A. Hotho, G. Stumme Proc. ISWC2008, LNCS 5318, 615-631 (2008) http://isiosf.isi.it/~cattuto/papers/cattuto_iswc2008.pdf

* “Evaluating Similarity Measures for Emergent Semantics of Social Tagging”

B. Markines, C. Cattuto, F. Menczer, D. Benz, A. Hotho, G. Stumme to appear in Proc. WWW2009 http://isiosf.isi.it/~cattuto/papers/socialsim_www2009.pdf

REFERENCE PERSON: Prof. Francesco Ricci, Francesco.Ricci@unibz.it


Gedas Adomavicius

TIME: Monday, February 16, 11:00AM-12:00PM

PLACE: A101 (Sernesi Building)

SPEAKER: Prof. Gedas Adomavicius (Carlson School of Management, University of Minnesota, Minneapolis, MN, USA)

TALK TITLE:

Overcoming Accuracy-Diversity Tradeoff in Recommender Systems

SPEAKER’S BIO:

Gediminas (Gedas) Adomavicius is currently an associate professor of

Information and Decision Sciences at the Carlson School of Management,

University of Minnesota. He received the PhD degree in Computer Science

from New York University. His research focuses on personalization

technologies and recommender systems, data mining and knowledge

discovery, and complex electronic market mechanisms, such as

combinatorial and multiattribute auctions. His research has been

published in leading Computer Science (CS) and Information Systems (IS)

journals, including IEEE Transactions on Knowledge and Data Engineering,

ACM Transactions on Information Systems, Data Mining and Knowledge

Discovery, Information Systems Research, INFORMS Journal on Computing,

and IEEE Intelligent Systems. He serves on the editorial boards of

Information Systems Research and INFORMS Journal on Computing, and has

also served on program committees of numerous CS and IS conferences. In

2006, he received the U.S.

National Science Foundation CAREER award for his research on

personalization technologies.

SPEAKER’S HOMEPAGE:

http://ids.csom.umn.edu/faculty/gedas

TALK ABSTRACT:

Collaborative filtering and, more generally, recommender systems

represent an increasingly popular and important set of personalization

technologies that help people navigate through the vast amounts of

information. The performance of recommender systems can be evaluated

along several dimensions, such as the accuracy of recommendations for

each user and the diversity of recommendations across different users.

Intuitively, there is a tradeoff between accuracy and diversity, because

high accuracy may often be obtained by safely recommending to users the

most popular (”bestselling”) items, which can lead to the reduction in

recommendation diversity, i.e., less personalized recommendations. And

conversely, higher diversity can be achieved by trying to uncover and

recommend highly idiosyncratic/personalized items for each user, which

are inherently more difficult to predict and, thus, may lead to a

decrease in recommendation accuracy. This talk explores several

approaches that can improve both the accuracy and diversity of

recommendations obtained from different collaborative filtering

techniques, and provides empirical results based on several real-world

movie rating datasets.

REFERENCE PERSON: Prof. Francesco Ricci, Francesco.Ricci@unibz.it


Arta Dilo

TIME: Monday, January 26, 11:00AM-12:00PM

PLACE: E221 (Sernesi Building)

SPEAKER: Prof. Arta Dilo (Technical University of Delft, the Netherlands)

TALK TITLE:

Data structures for progressive transfer of 2D spatial data in a client-server environment

SPEAKER’S BIO:

I am currently a postdoc researcher at GIS-technology section, Technical University of Delft, the Netherlands. I hold a PhD in Geoinformatics from the International Institute for Geo-Information Science and Earth Observation (ITC) and Wageningen University, the Netherlands, an MSc degree from ITC in GeoInformatics, and a BSc degree from the University of Tirana, Albania, in Applied Mathematics.

My recent and current research include data modelling; spatial databases; formal aspects of representation of (spatial) information, related implementation issues; formal reasoning, and data analysis.

SPEAKER’S HOMEPAGE:

http://www.gdmc.nl/arta/

TALK ABSTRACT:

In recent years Internet has become an important source for digital maps. The classical problem of (analogue) map generalisation is translated to needs for automatic generalisation of (digital) spatial data. Offering a digital map to clients over the internet, requires generation & transmission of different scales of the map, i.e. generalisation results for different levels of detail (LoD). In our approach we perform an off-line generalisation, and store its results in a collection of data structures, called tGAP (topological Generalised Area Partition), which can be queried to generate maps at any required LoD. In this talk I will describe the tGAP data structures, the algorithm that performs the generalisation and fills the tGAP, and retrieving data from tGAP in order to create maps at different LoDs. An important characteristic of tGAP is that it stores changes between two consecutive LoDs, thus allowing to identify the difference between maps of two different LoDs as an accumulation of such changes. This makes possible the refinement of a displayed map at a client, by sending progressively the changes. The use of tGAP for progressive transfer will also be explained in this talk.

REFERENCE PERSON: Prof. Michael Böhlen, boehlen@inf.unibz.it


Arturas Mazeika

TIME: Friday, January 23, 3:00PM-4:00PM

PLACE: A101 (Sernesi Building)

SPEAKER: Prof. Arturas Mazeika (Max-Planck-Institut füer Informatik, Saarbrücken, Germany)

TALK TITLE:

Bachelor: Minimizing Stochastic Coherence of Web Archives

SPEAKER’S BIO:

Arturas Mazeika is a researcher in the Database and Information Systems

Group at Max-Planck-Institut für Informatik, Germany. He holds a MSc

from the Department of Mathematics and Computer Science at Vilnius

University, and PhD at the Departments of Computer Science and

Communication at Aalborg University. His research interests include

visual data mining, data mining, and databases. He is a main developer

of an interactive 3D visual data mining system that has been used to

analyze a variety of datasets. He publishes regularly in the main data

mining and database outlets.

SPEAKER’S HOMEPAGE:

http://www.mpi-inf.mpg.de/~amazeika/

TALK ABSTRACT:

The content on Internet undergoes very rapid changes. Studies suggest

that over 60% of Internet changes every year. National libraries as

well as other non-profit organizations started to archive and preserve

the content of the web from disappearing. A typical unit of archival

usually is a whole site (a few thousand of pages), typically it takes a

few days to politely crawl the web site, and often the web site

undergoes a number of changes during the crawl time. Understanding how

much the web site has changed during the archive and what is the

incoherence level of the archive is the focus of the talk.

In this talk we will present a stochastic approach to measure the

incoherence of a web archive. We introduce Bachelor-offline, the optimal

solution under the assumption of the Poisson model and complete

knowledge of the pages of the web site in advance. Bachelor-offline

schedules fast changing pages in the middle of the crawl and the static

pages in the beginning and at the end of the crawl. We introduce

Bachelor-online, the online version of Bachelor-offline strategy where

the information about the pages of the web side are discovered during

the crawl. A thorough analytical and experimental investigation of the

Bachelor-offline and Bachelor-online strategies shows that our solutions

outperform other competitors.

The talk presents the latest findings of the work in progress. This work

is done together with Dimitar Denev, Marc Spaniol, and Gerhard Weikum.

REFERENCE PERSON: Prof. Michael Böhlen, boehlen@inf.unibz.it


Alejandro Vaisman

TIME: Wednesday, January 21, 3:00PM-4:00PM

PLACE: A101 (Sernesi Building)

SPEAKER: Prof. Alejandro Vaisman (Hasselt University, Diepenbeek, Belgium)

TALK TITLE:

Constraint Evaluation in Categorical Sequential Pattern Mining

SPEAKER’S BIO:

Alejandro Vaisman was born in Buenos Aires, Argentina. He received Civil Engineer and Bachelor degrees in Computer Science, and a PhD in Computer Science (under the supervision of Alberto O. Mendelzon) from the University of Buenos Aires (UBA). He has been a post-doctoral fellow at the University of Toronto in 2003. He is a Professor at the UBA since 1994. He was an invited professor at the Universidad Politecnica de Madrid in 1997, and visiting researcher at the Universities of Toronto and Hasselt, and Universidad de Chile. His research interests are in the field of databases, particularly in OLAP, Datawarehousing, Data Mining, XML and the Semantic Web, and Geographic Information Systems. He manages the project “Using OLAP Techniques in Geographical Information Systems”, funded by the Argentinian Scientific Agency in 2006. He is currently on leave from UBA, with the Theoretical Computer Science Group at the University of Hasselt, Beligum.

SPEAKER’S HOMEPAGE:

http://www.cs.toronto.edu/~avaisman/

TALK ABSTRACT:

The classic Generalized Sequential Patterns (GSP) algorithm returns all frequent sequences present in a database. However, usually a few ones are interesting from a user’s point of view. Thus, post-processing tasks are required in order to discard uninteresting sequences. To avoid this drawback, languages based on regular expressions (RE) were proposed to restrict frequent sequences to the ones that satisfy user-specified constraints.

In all of these languages, REs are applied over items, limiting their applicability in complex real-world situations. We propose a much powerful language, based on regular expressions, denoted RE-SPaM, where the basic elements are constraints defined over the (temporal and non-temporal) attributes of the items to be mined. Expressions in this language may include attributes, functions over attributes, and variables. We have applied this approach to trajectory database mining, although it is general enough to be used in other settings.

In this talk we will sketch the syntax and semantics of RE-SPaM, and present a comprehensive set of examples to illustrate its expressive power. We will also show how the language can be used for early pruning of sequences during the mining process, and give an overview of the algorithm we developed for this, and the pruning techniques aimed at reducing the need to access the database. Finally, we will comment on our experimental results, and, time allowing, present a demo of the language.

REFERENCE PERSON: Prof. Michael Böhlen, boehlen@inf.unibz.it


Cosimo Palmisano

TIME: Tuesday, December 9, 2:00PM-3:00PM

PLACE: A101 (Sernesi Building)

SPEAKER: Dr. Cosimo Palmisano (Fiat S.p.A.)

TALK TITLE:

A Data Mining Strategy For Targeted Sales Actions: A Case Study

SPEAKER’S BIO:

Cosimo Palmisano received the Master’s degree in electronic engineering and the PhD degree in management engineering from the Polytechnic of Bari, Italy. He is a senior business data analyst consultant in the Customer Insight Department, FIAT Group Automobiles. He has been a visiting scholar in the Information and Operations Management System Department, Stern School of Business, New York University. His current research interests include e-business models, customer relationship management, recommender systems, and utilizing contextual information in personalization problems. He has published numerous papers in various information systems and management international conferences and journal. He has been involved in several industrial research projects related to the above-mentioned research fields. He is a member of the IEEE and the IEEE Computer Society and reviewer for IEEE Transactions on Knowledge and Data Engineering Journal.

SPEAKER’S HOMEPAGE:

TALK ABSTRACT:

This seminar provides an insight about how to transform a data mining activity in a marketing strategy. It describes why and how to build a strategy not only to model the business problem but also for communicating the results to the internal clients and building up a ready to use platform for helping brands and managers to use data mining outputs and evaluate the performance of CRM activities in terms of Return of Investment.

REFERENCE PERSON: Prof. Francesco Ricci, Francesco.Ricci@unibz.it


Paolo Cremonesi

TIME: Tuesday, November 25, 2:00PM-3:00PM

PLACE: A101 (Sernesi Building)

SPEAKER: Prof. Paolo Cremonesi (Politecnico di Milano)

TALK TITLE:

A Large-Scale Recommender System for an IP Television Service Provider

SPEAKER’S BIO:

Paolo Cremonesi is associate professor at the “Politecnico di Milano”

where he teaches courses on “Computer System Architectures” and “Digital

and Internet Television”. During his 15 years with the “Politecnico di

Milano”, he led a large number of projects on the performance assessment

and optimization of large-scale datacenters as well as a new research

team devoted to recommender systems. He’s author of more than 40

scientific publications and books concerning distributed systems,

performance evaluation and capacity planning.

He holds a MS in Aerospace Engineering and a PhD in Computer Science,

both from the “Politecnico di Milano”.

SPEAKER’S HOMEPAGE:

TALK ABSTRACT:

This talk describes the development and integration of a recommender

system into the production environment of one of the largest European IP

Television

(IPTV) providers. The recommender system uses both collaborative and

content-based algorithms, suitable tailored to the requirements of an

IPTV architecture. The recommender system selects the proper algorithm

depending on the context.

The integration of the recommender system has faced with several

problems peculiar of an IPTV service, such as the presence of a large

catalog of thousands of multimedia contents, a customer base of more

then 200′000 users, the need to satisfy real-time constraints, and the

limitations of the user interface (television).

The recommender system is actually providing, on average, 30′000

recommendations/day, with peaks of almost 120 recommendations/minute

during peak hours. Almost 10% of the recommended items are purchased by

the users, confirming the accuracy of the recommendations estimated

during the testing phase.

REFERENCE PERSON: Prof. Francesco Ricci, Francesco.Ricci@unibz.it


Thomas Roth-Berghofer

TIME: Monday, November 3, 11:00AM-12:00PM

PLACE: A101 (Sernesi Building)

SPEAKER: Dr. Thomas Roth-Berghofer (DFKI GmbH / TU Kaiserslautern, Germany)

TALK TITLE:

A Unifying View on Explanation-aware Computing – Supporting the Use of Complex Information Systems

SPEAKER’S BIO:

Thomas Roth-Berghofer is Senior Researcher at the German Research Center for Artificial Intelligence DFKI GmbH, and Lecturer at the University of Kaiserslautern on such topics as Case-Based Reasoning and the Semantic Web. His main research interests are in complex (personal) information systems with explanation capabilities. Dr. Roth-Berghofer initiated several workshop series on: Explanation-aware Computing (ExaCt), Modelling and Retrieval of Reasoning in Context (MRC), and Philosophy and Informatics (WSPI). He organised several conferences and is general Chair of the seventh International and Interdisciplinary Conference on Modelling and Using Context (CONTEXT 2009), in Prato, Italy. Please visit his homepage and his blog for more information about him: http://thomas.roth-berghofer.de.

SPEAKER’S HOMEPAGE:

http://www.dfki.uni-kl.de/~roth

TALK ABSTRACT:

Explanation has some history in knowledge-based systems. Explanation, trust, and transparency are concepts that are strongly connected. One trusts a (software) system much more if it is able to explain what it is doing and why, and, thus, can “prove” its trustworthiness to its user. Explanations are part of human understanding and communication processes, and, therefore, should be incorporated into system interactions in order to improve decision-making processes. As information systems grow more and more complex, computer support in form of advanced explanation capabilities is needed. In the recent past the topic of explanation became of interest in general computer science under the name explanation-aware computing and is addressed in a series of workshops de-voted to that topic, mainly due to the insight that explanation is part of a communication process. This new view opens the road to considering more kinds of explanations than in the past. In this talk I will discuss conceptual and foundational issues of explanation-aware computing, considering explanations as a special kind of information that has certain utilities associated with it. Among the concepts connected with information are correctness and understanding, which can be subsumed under utility. That an explanation is some kind of information is in principle quite clear and may be even seen as trivial, but a systematic treatment of this as-pect is mainly missing. I will also show some prototypical implementations from different projects.

REFERENCE PERSON: Prof. Francesco Ricci, Francesco.Ricci@unibz.it


Daniele Quercia

TIME: Wednesday, October 29, 11:30AM-12:30PM

PLACE: A101 (Sernesi Building)

SPEAKER: Daniele Quercia, University College London, UK

TALK TITLE:

Mobile Content-Sharing Applications

SPEAKER’S BIO:

Daniele Quercia is a Microsoft Research Cambridge PhD Scholar and is finishing his PhD at the Computer Science department of the University College London, working in the Networks Research Group. He has studied new ways for co-located mobile users to share digital content (e.g., pictures, news). He has also been interested in commercializing new technologies while being: (1) a student of MBA “technology electives” at London Business School (consistently ranked as one of the top business schools in the world); and (2) a member of the “UCL Centre for Security and Crime Science”, which offers holistic security solutions to organisations from industry, commerce and government. Before commencing a PhD in the department, he was a Research Fellow in the area of mobile computing. He studied at Politecnico di Torino (PoliTO), Italy, for his Master of Science in Computer Engineering. As a recipient of international scholarship awards (while at PoliTO), he was a visiting student at Universitaet Karlsruhe, Germany, and University of Illinois, Chicago, USA.

SPEAKER’S HOMEPAGE:

http://www.cs.ucl.ac.uk/staff/d.quercia

TALK ABSTRACT:

Using mobile phones, people may create and distribute different types of digital content (e.g., photos, videos). Then, to avoid content overload, they need to locate quality content. They may do so by having their phones exchange ratings with each other. I will talk about:

  1. How mobile phones store ratings in a secure way Mobile networks suffer from content providers who tweak ratings to their own advantage. I will present a new decentralized mechanism (dubbed MobiRate) with which mobiles store recommendations in (local) tamper-evident tables and check the integrity of those tables through a gossiping protocol.Relying on my evaluation upon real mobility and social network data, we will see that MobiRate: (1) considerably reduces the impact of “tweaked” ratings; and (2) runs on mobile phones at little computational and communication costs.
  2. How mobile phones use those ratings for locating quality content I will talk about a new algorithm with which mobiles set their trust from third-party recommendations. The idea is that each mobile phone organizes ratings of content producers that it knows and trusts in a graph (called “Web of Trust”). It then uses a graph-based learning technique to form opinions about content producers with whom it has never interacted before. We will see that this algorithm shows high predictive accuracy on a large real-world dataset, and that, in contrast to existing approaches, the algorithm is fully decentralized and runs reasonably fast.

REFERENCE PERSON: Prof. Francesco Ricci, Francesco.Ricci@unibz.it


Christian Müller

TIME: Monday, October 13, 3PM-4PM

PLACE: A101 (Sernesi Building)

SPEAKER: Prof. Christian Müller, German Research Center for Artificial Intelligence (DFKI)

TALK TITLE:

Recommendations Based on Speech Classification (and examples of what recommender systems can learn from signal processing)

SPEAKER’S BIO:

Dr. Christian Müller is a senior researcher at the German Research Center for Artificial Intelligence (DFKI) in the Intelligent User Interfaces (IUI) department. Since June 2006, he has been a visiting researcher at the International Computer Science Institute (ICSI) in Berkeley, California. During his stay at ICSI, he was able to acquire substantial post-doctoral experience in human language technology. He earned a Ph.D. in Computer Science at Saarland University in January 2006. His dissertation “Zweistufige kontextsensitive Sprecherklassifikation am Beispiel von Alter und Geschlecht” (Two- layered Context-Sensitive Speaker Classification on the Example of Age and Gender) was supervised by Prof. Wolfgang Wahlster and graded Magna Cum Laude. From 1994 to 2001, he studied Computational Linguistics at Saarland University where he graduated with distinction. His research interests are inter alia speech and human language technology in general; more specifically: speaker classification (age, gender, cognitive load, emotions, etc), speaker recognition, language recognition, acoustic event detection.

SPEAKER’S HOMEPAGE:

http://w5.cs.uni-sb.de/~cmueller/m3i/php/website2.php

TALK ABSTRACT:

I this talk, I am going to present results from my work on age group specific product recommendations which was a joint project with the Deutsche Telekom. I am going to introduce briefly the idea of adapting a spoken dialog system according to the age (and gender) of the user.

I will focus, though, on the methods of recognizing the speaker age on the basis of the voice and speaking behaviour. I am also going to report on my work on language, speaker, and acoustic event recognition which evolved insights in several aspects of machine learning which I believe are of interest for the recommender system community. Particularly I am going to talk about the rank normalization method and combining generative with discriminative methods in the so called supervector approach.

REFERENCE PERSON: Prof. Francesco Ricci, Francesco.Ricci@unibz.it


Curtis Dyreson

TIME: Friday, September 5, 3PM-4PM

PLACE: D003 (Sernesi Building)

SPEAKER: Prof. Curtis Dyreson, Utah State University, USA

TALK TITLE:

About Closeness in XML and Supporting Proscriptive Metadata in an XML DBMS

SPEAKER’S BIO:

I am an assistant professor in the Department of Computer Science at Utah State University. I’m also the ACM SIGMOD DiSC Editor, the ACM SIGMOD Anthology Editor, the Information Director for ACM Transactions on Database Systems, and serve on the SIGMOD Executive Committee. I teach and research mostly in the area of database systems. My interests include temporal databases, native XML databases, data cubes, and providing support for proscriptive metadata. Prior to coming to USU I was

a professor at Washington State University, James Cook University, Aalborg University, and Bond University. I am a graduate of New College and obtained my Ph.D. from the University of Arizona.

SPEAKER’S HOMEPAGE:

http://www.cs.usu.edu/~cdyreson/

TALK ABSTRACT:

This talk consists of two separate talks on XML in the context of database systems. The first talk is about improving queries in hierarchies. Hierarchies are common in computer systems, from the XML data model in database management systems to file directories in operating systems. When a single hierarchy is imposed on data, a query becomes “brittle” in the sense that it might fail to produce the desired result when executed on the same data organized in a different hierarchy, when the hierarchy evolves to a slightly different hierarchy, or when a user misunderstands the hierarchy when constructing a query. A fixed hierarchy also makes it more difficult to integrate heterogeneous data sources since each source could organize similar data in a different hierarchy. To address these concerns we propose tracking “closest relationships” in an XML data collection. These relationships can be used to create a common representation for data from different hierarchies and can be used to

restructure data to any hierarchy. The second talk presents MetaXQuery. MetaXQuery is a language or querying data enhanced with metadata. I’ll discuss how we implemented MetaXQuery in eXist by eusing eXist’s indexes and query evaluation engine.

REFERENCE PERSON: Prof. Michael Boehlen, boehlen@inf.unibz.it