Combinaison des caractéristiques des termes pour l'extension de requêtes en recherche d'information dans les documents semi-structurés

In traditional Information Retrieval (IR), users often express their needs using simple keywords, because this is the most simple way to query systems. Queries composed of simple keyword terms are also very used in structured IR (i.e. IR on structured documents like XML documents), since they do not require any knowledge of the documents structure. However, such queries are often not sufficient to describe precisely users’ needs. In this paper, we propose a new approach of Structured Relevance Feedback on queries composed of simple keywords terms. This approach allows on the one hand to enrich the initial query according to relevance judgements, and on the other hand, to express a degree of preference for each keyword, by weighting them. Our approach is based on a combination of term features in relevant elements. Our proposition is evaluated thanks to the INEX evaluation campaign and results show the interest of our methods. MOTS-CLÉS : reformulation, document XML, contexte, contenu, RI.