FASP: Fuzzy Answer Set Programming


The goal of the FASP-project is the development of fuzzy answer set programming as a generalization of the existing answer set programming. It is expected that the combination with fuzzy techniques will greatly expand the application potential of answer set programming, in particular in the area of the semantic web. For example, it will become possible to express the (reasons for) trust and distrust between agents, to model the appreciation of possibly conflicting information obtained from various sources, to allow for less important rules to be satisfied only to a certain extent etc.

This project is supported by the Fund for Scientific Research-Flanders. It is carried out in close cooperation with the Theoretical Computer Science Group at the Free University of Brussels.

Project website

AutoWeb: Ontology Learning


The goal of the AutoWeb-project is to endow web agents with the thinking ability that allows them to maintain their ontology, by continuously questioning the knowledge contained in it and comparing this knowledge with answers retrieved from the semantic web. We develop the agents' skills w.r.t. deducing implicit knowledge, giving explanations, and trusting or distrusting various sources.

This project is supported by the Special Research Fund of Ghent University. It is carried out in close cooperation with the Laboratory for Automatic Information Retrieval at Hogeschool Gent.

Trust Networks for Recommender Systems


Recommender systems (RSs) are systems designed to suggest items (movies, music, books, news, Web pages, etc.) to users who might be interested in them, given some information about the user's profile. Although there are some good recommender techniques, it's still very difficult to generate highly qualitative recommendations, especially when dealing with new users. We can improve the quality and the amount of the delivered recommendations by using a trust network (a network in which the users are connected by trust scores).
Therefore, the goal of this project is the creation of a framework that can represent partial trust, distrust and ignorance, that contains appropriate operators (i.e. propagation and aggregation) to work with those trust couples, and ultimately, that can be incorporated in RSs. L-fuzzy set theory will be used to model the trust network, which allows us to approach the human interpretations of trust and distrust as closely as possible.

This project is supported by the Institute for the Promotion of Innovation through Science and Technology in Flanders (IWT-Vlaanderen) through a PhD-scholarship.

Personalizing Access to Large Data Volumes


With the advent of large-scale online applications, personalization is gaining momentum on the Web as a means of challenging the information overload, as well as of understanding, and catering to, the needs of individuals or groups. Rather than maintaining uniform, one-fits-all access channels to online resources, applications like recommender systems and personalized search engines aim to diversify and individualize users' view of data collections, producing information in the right format, at the right time, to the right user. In the framework of e-commerce for example, this requires a creative and intelligent handling of earlier sales' transactions, clickstream data, seller-buyer interaction, analogies between customers, etc. to be able to offer "user-tailored" recommendations to each customer. Taking into account the heterogeneous, dynamic and often incomplete or conflicting nature of these various sources of information, what is indispensable is a robust knowledge mining and processing mechanism; unsurprisingly, personalization has been described as a true "killer application" for artificial intelligence. The specific aim of our project is to exploit to the fullest extent the potential of granular and "soft" computing methods so as to contribute to a more user-friendly, more productive and more homogeneous information society, as well as to a more satisfying Web experience.

This project is supported by the Fund for Scientific Research-Flanders through a postdoctoral mandate.

Intelligent Question Answering


Question answering systems (QA-systems) are information retrieval systems that are able to answer natural language questions using large document collections, such as the web, as their primary knowledge base. Most QA-systems rely on the assumption that an answer to a question, if present in the collection, can be found in a single sentence. While this seems reasonable for simple factoid questions, it severely restricts the potential of a QA-system to handle more complex question classes. This project aims at improving the reasoning capabilities of QA-systems in order to obtain focused intelligent QA-systems in which answers can be deduced by combining information from different sources. Qualitative reasoning and fuzzy set theory are used as the key techniques to bridge the gap between natural language descriptions and the formal representation of knowledge that is required for automated reasoning.

This project is supported by the Fund for Scientific Research-Flanders through a PhD-scholarship.
Website  Computational Web Intelligence TeamComputational Web Intelligence