Towards a synthetic analysis of user's information need for more effective personalized filtering services

The consideration of underlying analysis of user's information need is a key requirement in an intelligent filtering environment. However, the majority of current approaches to filtering are relevance-oriented, rather than user-oriented. This is partly because they are issued from fields that have somewhat different perspectives from that of information filtering, but also because of the difficulty of understanding and measuring user's motivations and the way in which the user expects the system to respond. This paper presents an original approach to information analysis and filtering inspired by the novelty detection theory. As well as being able to accurately learn user's information need, the approach has an analytical capacity for better understanding user's need. It provides a new way of looking at user's need in terms of precise, broad, and contradictory profile-contributing criteria. These criteria go on to estimate the relative importance the user might attach to precision and recall. The filtering threshold is then adjusted taking into account this knowledge about user's need. Experimental results on the standard Reuters-21578 collection prove the effectiveness of the approach and confirm the potential usefulness of adapting the filtering results according to the knowledge acquired about user's need.