基于知识图注意网络的个性化推荐算法

现有的大多数利用知识图谱的推荐算法在探索用户的潜在偏好时没有有效解决知识图谱中存在的不相关实体的问题,导致推荐结果准确率不高。针对这一问题,提出了基于知识图谱和图注意网络的推荐算法KG-GAT(knowledge graph and graph attention network)。该算法将知识图谱作为辅助信息,在图注意网络中使用分层注意力机制嵌入与实体相关的近邻实体的信息来重新定义实体的嵌入,得到更有效的用户和项目的潜在表示,生成更精确的top-N推荐列表,并带来了可解释性。最后利用两个公开数据集将所提算法和其他算法进行实验对比,得出所提算法KG-GAT能够有效解决沿着知识图谱中的关系探索用户的潜在偏好时存在的不相关实体的问题。

Personalized recommendation algorithm based on knowledge graph attention network

Most existing recommendation algorithms using knowledge graph could not solve the problem of irrelevant entities in the knowledge graph when detecting the potential preferences of user,resulting in low accuracy of recommendations.To address this problem,this paper proposed a recommendation algorithm KG-GAT based on the knowledge graph and graph attention network.This algorithm used the knowledge graph as auxiliary information,and the graph attention network redefined the embedding of the entity by incorporating the information about the neighboring entities related to the entity in a layered attention mechanism.It obtained a more effective potential representation of users and items,generated a more accurate top-N re-commendation list,and provided interpretability as well.Compared with other algorithms,the experimental results on two open datasets show that KG-GAT can effectively solve the problem of unrelated entities when detecting the potential preferences of user along with the relationships in the knowledge graph.