A Restricted Boltzmann Machine-based Recommender System For Alleviating Sparsity Issues

With the explosive growth of the Internet and the Web, assisting users and facilitate their access to resources that might be of their interest and that are adapted to their personal needs is a tedious task. Efficient management of large amounts of information becomes an increasingly significant challenge. Hence, recommender systems have proved, in recent years, to be a valuable asset to dealing with the problem of information overload by assisting the users and providing them with more effective access to information. To this end, these systems must be able to predict users' interests based on their prior feedback. However, sparsity issues arise when necessary transactional information is not available for inferring users and items similarities, which deteriorate the quality and accuracy of the recommender system. To fill these gaps, we propose in this paper a Restricted Boltzmann Machine-based model to learn hidden factors and reconstruct sparse input rating data. Experimental results show that our proposed approach can effectively deal with data sparsity in MovieLens dataset, containing a massive amount of scarce information.