Abstract
Sparsity is a serious problem of collaborative filtering (CF) that has a considerable effect on recommendation quality. Contextual information is introduced in traditional recommendation systems besides users' and items' information to overcome this problem. Several research works proved that incorporating contextual information may increase sparse data. For this, data-mining techniques are among the most effective solutions that have been used in context-aware recommendation systems to handle the sparsity problem. This paper proposes the combination of a new context-user-based similarity collaborative filtering recommendation technique with data mining techniques, as a solution to this problem and develops a novel recommendation system: Rule-based Context-aware Recommender System (R_CARS). R_ CARS is experimented introducing four rule-based algorithms: JRip, PART, J48 and RandomForest, on four different datasets: DePaulMovie, InCarMusic, Restaurant and LDOS_CoMoDa and compared with the state-of-the-art models. The results of the experiment show that weighting the rating-based similarity with context and combining it with a rule-based technique can overcome the sparsity problem and significantly improve the accuracy of recommendation compared to the stateof-the-art models.