Personalized recommendation system based on knowledge embedding and historical behavior

Collaborative filtering (CF) usually suffers from limited performance in recommendation systems due to the sparsity of user-item interactions and cold start problems. To address these issues, auxiliary information from knowledge graphs, such as social networks and item properties, is typically used to boost performance. The current recommended algorithms based on knowledge graphs fail to utilize rich semantic associations. In this paper, we regard knowledge graphs as heterogeneous networks to add auxiliary information, propose a recommendation system with unified embeddings of behavior and knowledge features, and mine user preferences from their historical behavior and knowledge graphs to provide more accurate and diverse recommendations to the users. Our proposed ReBKC shows a significant improvement on three datasets compared to state-of-the-art methods. These results verify the effectiveness of learning short-term and long-term user preferences from their historical behavior and by integrating knowledge graphs to deeply identify user preferences.