Reliefs: un système d'inspiration cognitive pour le filtrage adaptatif de documents textuels

This paper deals with the description of a new adaptive filtering system called RELIEFS (for RELevance Information Fuzzy System). The main principles of this system draw their inspiration from cognitive models of information selection. More precisely, our research is based on the analysis of semantic memory models (knowledge access and knowledge organisation) and attentional models (information selection from the environment). We also propose to bridge a gap between these cognitive models and traditional information retrieval models. A new interpretation of relevance is given. The analysis leads us to extract simple mechanisms refering to the concepts of activation and spreading activation for a system which aims at selecting "relevant" information. These mechanisms have been implemented and successfully tested in the TREC9 adaptive filtering task. MOTS-CLÉS : filtrage adaptatif, pertinence, cognition.