Utilisation du réseau sémantique de l'UMLS pour la définition de types d'entités nommées médicales

Named Entities are important concepts, regarding Question-Answering (QA) systems. Nevertheless, Named Entities categories are usually defined in a very broad sense: date, geographical area, and so on. It should quite profitable, for medical QA systems, to benefit from Named Entities especially dedicated to medicine. The UMLS is an important terminological tool created and maintained by the NLM; it comes along with a hierarchical organization of medical concepts. This paper is an attempt to evaluate the French part of UMLS as a resource for a medical-specific Named Entity tagger. We have tagged a set of medical documents, and have shown, using statistical studies that strategies using these new tags in a QA context are to take in account the individual origin of each document. MOTS-CLÉS : UMLS, entités nommées, systèmes de Questions-Réponses, recherche d’information, réseau sémantique, étiquetage, structure thématique, analyse de données, analyse de corpus.