Construction et utilisation de contextes autour des noeuds d'un hypertexte pour la recherche d'information

We assume that the nodes of an hypertext are not always self-contained because when some information is made available in a hypertextual form, it is split in many nodes. Under this hypothesis the sole content of a node is not sufficient to index it in order to insert it in an information retrieval system. Using a clustering method, we have implemented and tested a context building method around the nodes of an hypertext. The similarity function between the nodes is based both on hypertext structural clues - the hyperlinks - and the textual content of the nodes. Our information retrieval system indexes both the nodes and their contexts. We use two level queries: the subject level and the context level.