Propositions pour la pondération des termes et l'évaluation de la pertinence des éléments en recherche d'information structurée

Structured Information Retrieval copes with a number of open issues. On e can cite the evaluation of elements relevance: the aim is to return to the user a ranked list of results. This evaluation is based on the term weighting scheme and on the model used for th matching of queries and elements. In this paper, we propose to explore some clues to answer to this problem. For example, we propose to introduce the element context at different g ranularity levels. Our experiments use the XFIRM system on the INEX 2005 test suite. Results sho w the importance of elements specificity for the considered metrics. Contextual relevance s however to have no impact on elements relevance, which contradicts prior results. It high lights contradictions on the relevance metrics used in the INEX evaluation campaign.