Abstract de la publi numéro 7706

In Information Retrieval (IR), Relevance Feedback (RF) is a technique used to improve the precision of results returned by the Information Retrieval System. It usually consists in enriching the initial query by adding terms extracted from documents judged relevant by the user. This technique is largely used in full text Information Retrieval. The issue we address in this paper, relates to the effectiveness of such an approach in semi-structured documents like XML. Indeed, as information, the question is how to use these sources of evidence to process Relevance Feedback. In this paper, we propose what we call a structure-oriented relevance feedback approach: our goal is to enrich the initial query (composed of simple keyword terms) by adding structural constraints. Experiments undertaken on the INEX (INitiative for the Evaluation of XML Retrieval) collection confirm that refining the initial query by adding structural constraints is effective in XML Retrieval.