一种融合知识图注意神经网络的推荐算法

KNOWLEDGE GRAPH ATTENTION NETWORK FOR RECOMMENDATION SYSTEMS

为了提高推荐算法的准确性和可解释性,通常会在推荐算法中添加并利用用户和项目的一些辅助信息。大量实验表明,在推荐算法中添加知识图谱作为辅助信息,通过挖掘实体之间的相关属性可以有效地获取项目之间的相关性,从而较大地提高推荐的性能。受到图注意神经网络和KGCN的启发,设计一个注意嵌入传播层来计算知识图谱中实体的邻居信息,以丰富项目表示。在三个真实的数据集上进行实验,结果分析表明,在电影和书籍推荐中该算法推荐性能最佳,在音乐推荐中也取得了较好的推荐效果。

In order to improve the accuracy and interpretability of the recommendation algorithm,some auxiliary information of users and items is usually added and used in the recommendation algorithm.A large number of experiments show that adding knowledge graphs as auxiliary information to the recommendation algorithm can effectively obtain the correlation between items by mining the relevant attributes between entities,thereby greatly improving the performance of the recommendation.Inspired by the graph attention neural network and KGCN,an attention embedding propagation layer was designed to calculate the neighbor information of entities in the knowledge graph to enrich project representation.Experiments were conducted on three real data sets.The analysis of the results shows that this algorithm has the best recommendation performance in movie and book recommendation;in music recommendation,it has also achieved a high recommendation effect.