Bidirectional Knowledge-Aware Attention Network over Knowledge Graph for Explainable Recommendation
Yanxia Lyu, Guorui Su, Jianghan Wang, Ye Xing · 2022
Now recommendation systems are introduced into various online applications to help users find the content they want from massive data. Although the recommendation method based on collaborative filtering can adapt to the change of recommendation scenarios, the recommendation heavily depends on interaction data, so the recommendations are seriously affected by data sparsity. To alleviate the above problem researchers introduce knowledge graph as side information into the recommendation system. By exploring the rich entity information and relation information in knowledge graph, we can enrich the representations of user and item and enhance interpretability of recommendation system. However, some recommendation methods only carry out unidirectional knowledge propagation when mining knowledge information, which makes it difficult to capture higher-order knowledge information when the knowledge graph is sparse. Meanwhile, most recommendation models do not make full use of the relations between users, items and entities to enhance the interpretability of recommendations. Based on the reasons above, we design a novel bidirectional knowledge-aware attention network framework for explainable recommendation named BKANE, which integrates interaction information and high-order knowledge information, completing the recommendation in an end-to-end manner. The experimental results on three real datasets show that BKANE is significantly better than the state-of-the-art baselines in terms of recommendation performance. Also, the graphical explanation form can provide developers and users with a reasonable explanation of the model and recommendations.