KPRLN: deep knowledge preference-aware reinforcement learning network for recommendation

Abstract User preference information plays an important role in knowledge graph-based recommender systems, which is reflected users having different preferences for each entity–relation pair the graph. Existing approaches have not modeled this fine-grained user feature well, as affecting performance of systems. In paper, we propose a deep preference-aware reinforcement learning network (KPRLN) recommendation, builds paths between user’s historical interaction items graph, learns features user–entity–relation and generates weighted graph with features. First, proposed hierarchical propagation path construction method to address problems pendant entity long exploration The expands outward form clusters centered on uses them represent starting target states learning. With iteration clusters, can better learn explore farther paths. Besides, design attention convolutional network, focuses more influential pairs, aggregate item higher order representations that contain Finally, extensive experiments two real-world datasets demonstrate our outperforms other state-of-the-art baselines.