Attention Model in Recommender Systems

Abstract

Information available online is already very huge and it is also constantly growing. Presenting relevant information or information that is of interest to the user is not as easy as it sounds. To help users find information relevant to their interest and help advise them in making a decision recommender systems, are there for the rescue. Recommender systems are already being used extensively across the entire internet by most popular websites, their utility simply cannot be ignored. User-defined tags are frequently used by recommender systems, however these tags will have the draw backs of being sparse, ambiguous and redundant. To cater for these issues first we represent user profiles with tags and then we apply topic modelling to discover these abstract tags. Attention models have proven their usefulness in image processing, their influence in deep learning has recently been highlighted by applying them to information retrieval and recommender systems. Next a neural attention based model is used to filter out the relevant information. Using these extracted features, users’ profiles are refreshed and are used to create a recommendation list. The experimental studies have shown the usefulness of the proposed algorithm and it preforms better than deep neural network based algorithm.