CKGAT: Collaborative Knowledge-Aware Graph Attention Network for Top-N Recommendation

Knowledge graph-based recommendation methods are a hot research topic in the field of recommender systems recent years. As mainstream knowledge method, propagation-based method captures users’ potential interests items by integrating representations entities and relations graph high-order connection patterns between to provide personalized recommendations. For example, collaborative knowledge-aware attentive network (CKAN) is typical state-of-the-art that combines user-item interactions associations graph, performs heterogeneous propagation generate multi-hop ripple sets, thereby capturing interests. However, existing methods, including CKAN, usually ignore complex sets do not distinguish importance different resulting inaccurate user being captured. Therefore, this paper proposes top-N named attention (CKGAT). Based on strategy, CKGAT uses extract topological proximity structures then learn entity representations, generating refined set embeddings. further an aggregator perform weighted aggregation embeddings, user/item initial original accurate embeddings item for Experimental results show CKGAT, overall, outperforms three baseline six terms accuracy, four representative diversity.