Attention-enhanced Knowledge-aware Cross Network for Recommender Systems

Knowledge graph-based recommender systems use collaborative filtering data and knowledge graph data to improve prediction accuracy. Aiming at the problem that the knowledge-aware recommendation algorithm cannot fully mine and utilize high-order information, an improved knowledge-aware recommendation algorithm (AKCN) was proposed, which combined the knowledge-aware attention under collaborative guidance to better distinguish the importance between different neighbors, and added a cross-compression unit in the embedding module to make full use of high-order information. Experimental results on book Crossing and Last.-FM datasets show that AKCN achieves a certain performance improvement compared with CKAN and KGAT.