Modèles de langue pour la détection d'opinions dans les blogs

This article describes an approach to search relevant documents to the query and expressing an opinion. To detect if a document is opinionated (i.e; contain subjective information), we suggest to compare it with sources of information that contain subjective information. The intuition behind it is the following one, a document having a strong similarity with sources of opinions have an opinion. To measure this similarity we used languages models. We model the document and the reference of opinions using languages models, we estimate then the closeness of these models. Several experiments were realized on collections stemming from TREC. We took the collection of TREC blog06 as collection of analysis and the collection IMDB as being the collection of reference. MOTS-CLÉS : Recherche d’information, blogs, détection d’opinions, modèle de langue