Research On Recommendation Methods Based On Knowledge Graph

With the rapid development of the internet and big data technology,personalized recommendation systems are becoming increasingly important in many fields.However,traditional recommendation methods may face issues such as data sparsity when dealing with large amounts of data.Knowledge graphs,which cover a large number of entities and relationships,have attracted widespread attention in the research of recommendation systems.Applying knowledge graphs to recommendation systems can alleviate data sparsity issues and better understand users’ potential interests by mining entities and relationships in the knowledge graph.Additionally,path-based knowledge graph recommendation methods can provide good recommendation explanations,making them suitable for a wide range of application scenarios.In this paper,we propose two hybrid recommendation methods based on knowledge graphs,which improve recommendation performance from the perspectives of knowledge graph path exploration and measuring the comprehensive path scores from users to items.(1)The hybrid recommendation method based on reinforcement learning and knowledge graphs improves recommendation performance by integrating reasonably embedded representations of entities and relationships in the knowledge graph into a reinforcement learning-based recommendation system.To recommend diverse items,we define an exploratory reward function to establish a Markov decision process model.We also introduce an action space pruning strategy to narrow down the inference space and explain the objective for policy gradient optimization.Experimental results show that the proposed method can improve recommendation performance and provide reasonable explanations.(2)The hybrid recommendation method based on entity preferences and knowledge graphs improves recommendation performance by mining different semantic paths between entities,defining an entity preference calculation function and a comprehensive path score function,and calculating the preference of entities on each path based on user preferences,thus obtaining the comprehensive score of the current path and all paths,i.e.,preference prediction.The hybrid recommendation method based on entity preferences and knowledge graphs can predict users’ preferences for items,thereby obtaining more accurate recommendation results.Moreover,this pathbased method can better explain the recommendation results,increasing user trust in the recommendation system.We conducted experiments on real-world datasets comparing our method with other methods,demonstrating the effectiveness of the proposed method in terms of recommendation performance.