Intégration de connaissances syntaxiques dans les modèles de langue pour la RI

In Information Retrieval (IR), statistic keyword based methods have reached a limit. This limit can only be cross by integrating, in large number outside source of knowledge, in IR system. Our work is based on the integration of the syntactic link between the terms produced by shallow parser. For that we consider the syntactic dependency of dependency tree produce by parser. For integrating these information in IR, we present the use of a language modeling approach. Language modeling approach theoretically framework is attractive as it can be adapted or in order to take into account new information. We present here the integration of the dependency relation in a language model. We evaluate this model on a part of the CLEF collection. The results show that the integration of dependency relation lowers the IR results. Consequently, knowing these results in the continuation of this work, we intend to integrate more semantic information instead of syntactic information. MOTS-CLÉS : recherche d’information, analyse syntaxique, dépendance syntaxique, modèle de langue