Research Projects

This project aims at developing new preference prediction technologies for building proactive recommender systems. Standard recommender systems work in pull mode: the user is supposed to ask (pull) for a recommendation and therefore is required to enter information about his preferences, needs and constraints. A new kind of recommender systems are instead pushing the user with recommendations, when the user is in a contextual state that the system considers as suitable for that suggestions. For instance, when a user traveling in a region and passing by an interesting archeological site the system may suggest to visit it if that matches the user profile learned by the system. The specific aim of this project is to develop techniques that can identify the best set of data that are required from a user to deliver relevant recommendations to that user and for improving the system performance in the future. These techniques rely on Active Learning methodologies and we will study their application to the preference data sets managed by a recommender system. The ultimate goal is to identify the information, to be asked to the user, that is more useful for the system to deliver relevant information but that the user is also capable to provide, hence minimizing the cost of the preference elicitation process for the user. We will develop a system prototype of a mobile service that will provide push based recommendations for places of interest and events in South Tirol. Funding 43.077 Euro. Funding Agency: Autonomous Province Bozen-Bolzano.

This projects aims at defining the next functional steps towards an information relevance platform to be used by a Deutsche Telekom spinoff, Yoochoose. It is aimed at comparing recommendation technologies through benchmarks. It is also aimed at defining a set of metadata as they represent a critical factor for providing recommendations. The specific objectives are: identification of contact points to other fields like information search and user profile management; tools, equipment and data sets for benchmarking recommendation results; best practice documentation about the influence of new functions, settings of the system and metadata quality; cooperation towards a metadata architecture plus best practices for Entertainment applications. In a second phase, we will design and test a system that can deliver music recommendations for a group of user and that will adapt the recommendations based on the contextual situation of the group. The reference scenario is in car entertainment; the passengers will be recommended with music that is adapted to the car location, i.e., to the nearby points of interest, the time, and the user mood. In particular, the specific objectives of this second phase are: analyze DT requirements related to the personalization of music tracks for a group of users and for location-based music services; identify appropriate techniques for addressing these context-aware requirements; design and develop appropriate algorithms based on previous results developed in FUB on context-aware recommendations; test the proposed techniques and prototypes on a set of data provided by DT. Funding: 51.353 Euro - Funding agency: Deutsche Telekom

Analytical Services for Medical Data Warehouse – MOBAS (January 2008 - December 2011)

In this project we focus on the exploitation of mobile and ubiquitous computing techniques in the hospital and eHealth scenarios. In particular, when a patient is visiting a day hospital for periodical examinations, analysis, and treatments, executes a set of activities ideally organized in well defined workflows. However, the hospital is a highly dynamic environment and for this reason the waiting times between the activities in the workflow and even their actual sequence are often scarcely predictable. Therefore, it is important to timely inform the patients about their next activity, where it takes place, and when it starts, but also inform the user about other important aspects of their therapy, such as side effects, or to monitor their psycological status. In this research project we aim at designing and implementing novel mobile services, which are integrated in the hospital information system, that support patients and clinicians in the day hospital scenario and their follow up at home. We will design a message-posting algorithm for keeping in touch with the patient that uses context–aware rules elicited from the clinicians, or mined by observing previous interactions, to decide the time and content of the guidance messages that are pushed to the patient’s device. More in general we want to enhance the effectiveness of the communication flow between the patients and the physicians by exploiring the usage of several interaction channel, including mobile devices, personalized web sites and large screens. Funding: 125.000 Euro - Funding agency: Autonomous province of Bozen-Bolzano.

Partners: Ciro Cattuto (ISI Foundation, Torino) prof. Manfred Mitterer (Merano hospital), EDP Progetti Bolzano, prof. Bernard Holzner (Innsbruck University).

Real-Time Recommendation Revision and Explanation for a Network of Mobile Users – ReRex (February 2009 - December 2010)

The aim of this project is to advance the state of the art in recommender systems developing an effective methodology for supporting users in context-dependent real-time revision of personalized and contextualized recommendations. We have focussed mostly on the methodological aspect of context-dependent recommendations, and on testing the proposed techniques on a mobile application. We have developed three basic techniques: context dependent item weighting, context-dependent similarity adaptation and item splitting, context dependent and real-time user interfaces adaptation. We have developed a methodology for acquiring in-contex ratings for items that has been adopted to acquire the data used in the mobile recommender (ReRex). The mobile application (ReRex) is a context-aware iPhone-based recommender system for tourists. The system was developed in three steps: first the relevance of several important contextual factors were measured; then in-context ratings were acquired for a population of users; and finally a mobile application (ReRex), running on an iPhone, was designed, implemented and tested in a live users experiment. This application provides context-aware recommendations, visualizes them, and offers explanations for the proposed POIs. The recommendation list is updated as any contextual factor changes, hence supporting the replanning of the visit. Funding: 25.000 Euro - Funding agency: FUB.

Partners: prof. Gedas Adomavicius (University of Minnesota, USA), Xavier Amatrian (Telefonica R&D), and prof. Robin Burke (DePaul University, USA).

Adaptive Data Processing and Analysis Techniques in eGovernment - ADAPTe (January 2007 - December 2009)

Conversational recommender systems support a structured human-computer interaction in order to assist online users in important online activities such as travel planning. In this project we have studied the effects and advantages of a novel recommendation methodology based on Machine Learning techniques (Reinforcement Learning) that allows conversational systems to autonomously improve an initial strategy in order to learn a new one that is more effective and efficient. We applied and tested our approach within a prototype of an online travel recommender system in collaboration with the Austrian Tourism portal (Austria.info). We have shown that the learned strategy adapts its actions to the served users and deviates from a rigid initial strategy. More importantly, we show that the optimal strategy is able to assist online tourists in acquiring their goals more efficiently than the initial strategy. It can be used by the system designer to understand the limitations of an existing interaction design and guide him in the adoption of a new one that is capable to improve customer relationship, the usage of their web site, and the conversion rate of their online users. Funding: 40.554 Euro - Funding agency: FUB

Partners: prof. Derek Bridge (University College Cork, Ireland) and prof. Wolfram Hoepken (Director of the Tourism Competence Center Austria, ECCA).

Nowadays, more or less the complete Austrian tourism offer is available online. However, different services are often only available on different platforms and the customer is not supported in composing an individual journey, consisting of different services, comparable to a package tour. A complex search and booking process is the consequence. The objective of this project is to analyze and design a process for a dynamic bundling of tourism services, taking into consideration the requirements of all involved parties (customer, supplier and intermediary). Here, the integration of supplier-initiated and customer-initiated product bundles into one bundling process as well as the personalization of the bundling process is of high importance.

The research activities conducted by FUB aim at designing, implementing and testing learning techniques for the adaptation of system behavior to user responses to system actions. System actions may be recommendations, requests for user preferences, or display of personalized information. To achieve these goals we shall first adjust and install recommender and search engine functionality on the main Austrian tourism web site. Then we shall design and initiate a series of experiments to analyze the users’ behavior vis-à-vis systematically varied system response. This will require the definition of a new model for interaction state representation and its implementation. Then we shall incorporate alternative methods of statistical learning into the test system. These methods will include Reinforcement Learning approaches to learning dialogue strategies and Bayesian methods to represent and reason on user preferences.

Partners: ECCA eTourism competence center Austria (Austria), TRC tourism research center Krems (Austria), Österreich Werbung (Austria), Free University of Bozen-Bolzano (Italy), WU Wien (Austria), Donau-Universität Krems (Austria), Europäische Reise-versicherung (Germany), Infoterm International Information Centre for Terminology (Austria), Invent GmgH (Germany), Raiffeisen Informatik RIT (Austria), Xploration GmbH (Austria).

European Tourist Destination Portal – http://etd.ec3.at - (April 2004 to October 2005)

This new initiative was aimed at designing and developing a unique access point to all European National Travel Organizations web sites. This new multilingual (En, Fr, Ge, It, Sp and Pt ) web site was intended to bring a better visibility to Europe as a unique and composite destination, and to provide a quicker search tool to select the most suitable information. The portal content is extracted from the (National Tourism Organization) NTOs web sites and data bases, and is integrated in the portal using the Harmonise semantic data-reconciliation tool. The user is supported by the portal trough a wide range of services, including, trip planning and product/service recommendation. The project was coordinated by EC3 (A), and the other partners was: TISCover (A), Siemens (A), Lixto (A) and ITC-irst (I). Overall budget 1.9 MEuro (eual to funding), and ITC-irst budget is 193.800 Euro. I participated to the project proposal preparation, the development of the project and I was member of the Advisory Board.

This project is a follow up of a previous project (Harmonise). The technologies developed in Harmonise, were here validated in 11 case studies in different European countries. This enabled us to further improve and engineer the technology and to design a viable business model for the service (data reconciliation and interoperability infrastructure). HarmoTen is funded by the EU, in the eTen programme. The project coordinator is EC3 (A), and the other partners are: IFITT (A), Umbria Region (I), Consorzio Pisa Ricerche (I), ITC-irst (I). The overall budget is 829.836 Euro. ITC-irst budget is 21.600 Euro. I participated to the preparation of the proposal and now I’m participating to the project as ITC-irst responsible.

This project was aimed at developing and validating a recommender system for tourist destination decision-making. The DieToRecs recommender system provides an interactive web-based tool capable of supporting the structured process of selecting a tourist destination and bundling a personalised trip plan. DieToRecs accesses and integrates, by using a mediator architecture, data managed by two existing tourist web portals (trentino.to tiscover.com). DieToRecs recommender was designed as an adaptive system, i.e., capable to adapt the dialogue process and the suggested products/services learning user's characteristics and preferences. System learning and adaptation was performed at each session level (short term) and collecting a memory of previous supported human/computer recommendation sessions. The dialogue is driven by an explicit model of the tourist destination selection. DieToRecs supports product aggregation for a given destination, letting the user to build a personalised dynamic package. Key technologies: XML, OLAP, Case-Based Reasoning, Adaptive User Interfaces, Web and Data Integration Architectures. I’m the Project Director and eCTRL is working in partnership with: TISCover A.G.(A); University of Vienna (A), University of Urbana Champaigh (USA), APT Trentino (I). Budget 1.5 Meuro. I prepared the project proposal and coordinated the whole project as principal investigator.

This was the strategic and foundational project of the Electronic Commerce and Tourism Research Laboratory. The objective was to create a multidisciplinary approach towards IT and Tourism (in partnership between ITC-irst and University of Trento) and build up in Trento an international point of reference for IT&T. In the short term we have developed a new generation of web enabled intelligent tools (recommendation technologies) and built some real working prototypes (web based travel destination decision aid system and a mobile recommender system for service selection). In partnership with the Azienda di Promozione Turistica del Trentino we have developed a recommender system for touristic products/services information selection and travel planning. Seven researchers worked on this project. A first system prototype, called NutKing, has been available since 2002. NutKing is accessible at http://nutking.ectrldev.com/nutking/ and now this technology has been integrated in a product (Trip@dvice) available from www.ectrlsolution.com. NutKing supports high user interactivity, exploiting intelligent query refinement and products scoring, learning preferences from past user interactions (collaborative filtering). A plan for economic exploitation of the technology is underway, and a new company is going to be established as a spin off. Budget 1.9 Meuro. I participated to the preparation of the project proposal and coordinated the ITC-irst unit as principal investigator.

Although standardization initiatives (based on XML) have a long history in the field of tourism, with different levels of adoption and usage, the intended broad harmonisation of electronic markets has never been reached, due to the lack of flexibility and extensibility. Harmonise set up a co-ordinated initiative at European level, which has brought: