Explainable Recommendation Task Based On Reinforcement Learning

With the development of information technology and the explosive growth of network information,people rely much on the Internet to obtain information.Recommendation systems help people easily get the content they want.Researchers have tried different methods to implement recommendation,and reinforcement learning is one of the methods used.We combined reinforcement learning and knowledge graph to recommend items to target users.The explanation of recommended items is realized through the reasoning path in KG.In existing researches,embedding-based and pathbased methods of KG are used separately.We combine the two to make full use of KG.Trans E is used to get entity and relation embeddings.Our method KGDQN combines knowledge graph and reinforcement learning to determine suitable recommendation items.We define reasoning path examples.The path examples are used as reasonable paths to explain recommended items.After obtaining reasoning path examples,we find reason paths from the target user to the recommended item.Put path into the KGDQN model,and return results of the recommended items and reasons.Experiments conducted on the Amazon datasets showed that KGDQN has good performances.We compared different reasoning path examples,and results showed that the shortest path and the meta-path connected by the three purchase relationships performed better.