Although rich knowledge resources (e.g. UMLS) exist in biomedical area, their use in medical information retrieval have only produced moderate success. Previous attempts have tried to recognize concepts from documents and queries using these resources and have performed concept-based IR, but this approach often led to degradations in retrieval effectiveness. We believe that the situation is not only due to the difficulty of identifying concepts from texts, but also to the simplistic use of concepts. In this presentation, we propose a method for medical IR taking into account both the inherent query structure and the semantic structure between concepts. Our preliminary experiments with query structure have shown promising results. We expect to gain in retrieval precision by considering semantic relations between concepts.

Présentation de sujet de thèse

Jury: Guy Lapalme (président), Jian-Yun Nie (directeur), François Major (membre)