Recherche sociale et personnalisée d'Information
Nawal Ould Amer · HAL (Le Centre pour la Communication Scientifique Directe) · 2020
A wide range of services and platforms make the user more and more interactive with the web. A lot of information that concerns both users and resources (documents, images, videos, comments, tweets, tags, etc.) is constantly generated. This information can be very useful in information retrieval tasks, for user modeling. However, classical information retrieval models do not integrate the social context of the user.Therefore, a lot of research has been interested in combining these two areas of information retrieval and social networks, which has given rise to models of social information retrieval and personalised social information retrieval.The extraction, analysis and representation of information about the social activities of users play an important role in the personalized information retrieval systems. Hence, it is crucial to create accurate user models and infer their interests from all this information.In this thesis, we investigate how to create a user profile using folksonomies. We study the problem of terms weighting. Specifically, how to estimate among all the user data, the useful information that can be used to represent his interests.In the first part of this thesis, we present a review of state-of-the-art research on information retrieval and personalized social information retrieval work.In the second part, we describe our two main contributions. The first contribution of this thesis lies in the definition of a user tag-based model, where these tags cover the topics of the documents to which they are associated. Our approach is distinguished by the integration of the document content into the estimation of user tag weights.The second contribution of this thesis concerns the definition of a new approach of user modeling based on documents. The particularity of this model is to use user tags to estimate the relevant document terms. The goal is to select only the terms that describe the document topics, which interest the user.The last part of the thesis is dedicated to the evaluation of our proposals. The results obtained are very encouraging and our approaches improve the performance of the IR systems.