Hybrid Recommender System for a Context Aware Recommendation in the Film Domain

Abstract

Recommendation systems aim to offer personalized help in discovering relevant content. Several approaches have been designed for providing better recommendations that satisfy users’ needs. Based on ratings, on content, or on knowledge, isolated recommendation techniques often lack some good properties of other methods. Hence, hybrid combinations are able to compensate for those differences. Furthermore, the information to include in the recommendation is most of the time limited to the set of ratings users assigned to the items. By including additional information on where and when the recommendation is taking place, can improve the overall performance. Nevertheless, combining all these features into one single model is rather a daunting task due to its complexity, and often is disregarded as it might require some degree of domain knowledge. We propose a recommender system based on a model that captures the human understanding of how to produce a personalized recommendation. Moreover, by including context information, we try to enhance the overall user’s experience. This system is able to produce recommendations even under uncertainty. Hence, we used an explicit model which is in fact a Bayesian network, that directly encodes the relationships between users’ preferences, item attributes, and context information. The final recommendation is obtained by a two stage process, a combination of two recommendation strategies that complement each other. Such model is the Contextual Hybrid Bayesian Model.