Comparaison des stratégies d'indexation pour les langues asiatiques

In information retrieval, Chinese and Japanese present many challenging problems. Unlike most European languages, the lack of explicit word boundaries represents one of the most important issue for indexing. For this reason, many works proposed different approaches to index documents or requests written in these languages. This article presents a comparison of the common indexing strategies. More precisely, we compared four strategies for Chinese (1-grams, 2-grams, 1& 2-grams and words), two for Japanese (2grams, and words) and three for Korean (words, 2-grams, and morphemes). Using test collections of the NTCIR-5, we evaluated these various approaches on nine retrieval schemes: two probabilistic models and seven vectorial models. MOTS-CLÉS : Recherche monolingue ; stratégies d’indexation ; langues chinoise, japonaise et coréenne.