Un système de recommandation social et sémantique sensible au contexte.

Abstract

The work presented in this manuscript is in the area of Context-Aware Recommender Systems (CARS) which aims to improve traditional recommendation systems (SR) by taking into account context information in predicting process. However, SR suffers from some challenges, such as cold start and data sparsity. New methods are proposed to overcome these problems. We propose in our work, three contributions. The first contribution aims to overcome the limitations of the Context-Aware Splitting Approach (CASA), which represents one of the most effective pre-filtering approaches of the context-aware recommender system. We propose to add mainly trust information as well as semantics to improve the quality of prediction. The second contribution aims in proposing a hybrid approach entitled: (Trust based Context aware Post Filtering Approach(TCPoFA)) , which belongs to contextual post-filtering approach. The results of the experiments reveal that these approaches improve the relevance of the recommendations and outperform other non-contextual approaches in the literature in terms of precision. We propose another approach: Contextual modeling approach based on semantics and trust (ST-CAMA), which combines trust and context information using contextual weighting. Also, in order to select only trusted neighbors who have interests common to the target item for the active user, we propose to build clusters based on semantic similarities enriched semantically via linked open data (LOD).