HI2Rec: Exploring Knowledge in Heterogeneous Information for Movie Recommendation

Online movies' recommender systems aim to address the information explosion of movies and make the personalized recommendation for users. Recently, knowledge graphs have been proven to be highly effective to recommender systems, because they are able to fuse various recommendation models and can handle the issues of data sparsity and cold start to improve recommendation performance. However, less consideration is given to the information about the user's properties than the item's properties in the existing knowledge graph recommendation methods, which leads to some limitations in the recommendation results. In this paper, we propose HI2Rec, which integrates multiple information to learn the user's and item's vector representations for top-N recommendation to address the above-mentioned issues. We extract the movie-related information from the Linked Open Data and then leverage the knowledge representation learning approach to embed this information as well as real-world datasets' information of recommender systems to a unified vector space. These vector representations are further calculated to generate a preliminary recommendation list. Finally, we utilize a collaborative filter approach to generate a precision recommendation list. The experimental results on the real-world datasets demonstrate that HI2Rec gives substantial performance improvements against the state-of-the-art recommendation models.