A concept-based approach for indexing documents in IR

This paper addresses two important problems related to the use of semantics in IR. The first one concerns the representation of document semantics and its proper use in retrieval. The second is the integration of semantic-based retrieval with "traditional" keywords-based retrieval. The proposed approach aims to represent the document content by the best semantic network called document semantic core in two main steps. The first step extracts concepts (mono and multiword) from a document, driven by external generalpurpose ontology, namely WordNet. The second step builds the best semantic network by achieving a global disambiguation of the extracted concepts regarding to the document. Thus, selected concepts senses represent the nodes of the semantic network while the similarity measure values between them represent the arcs. The resulted scored concepts senses are used for conceptual indexing in Information Retrieval. MOTS-CLÉS : Recherche d'Information, Représentation Sémantique de documents, mesures de similarité sémantique, indexation conceptuelle, ontologies, WordNet.