知识图谱增强的神经协同过滤推荐方法

Knowledge Graph Enhanced Neural Collaborative Filtering Recommendation Method

推荐系统已经广泛应用于各领域以处理信息过载问题,但传统方法面临着数据稀疏的挑战,且使用矩阵分解也不能很好的捕获抽象的非线性交互.考虑到知识图谱可以提供丰富的边信息,文中提出一种知识图谱增强的神经协同过滤推荐方法.首先获取项目相关的元数据,将其构建为知识图谱,并利用表示学习方法获取图谱中的语义知识;其次,利用结合注意力的邻域传播机制获取图谱中的结构知识,以此增强项目表示;最后将得到的用户和项目表示送入矩阵分解与神经网络中进行推荐.在公开数据集MovieLens上的实验结果表明,该模型能够有效提升推荐结果的准确性.

Recommendation system has been widely used in various fields to deal with the problem of information overload, but traditional methods are faced with the challenge of data sparseness, and matrix factorization cannot well capture abstract nonlinear interactions. Considering that the knowledge graph can provide rich side information, this paper proposes a knowledge graph enhanced neural collaborative filtering recommendation model. First, crawl the item-related meta-data and build them as a knowledge graph, then use the representation learning method to obtain the semantic knowledge in the graph. Secondly, attention combined neighbor propagation mechanism is used to obtain the structural knowledge in the graph, so as to enhance the item representation. Finally, the obtained user and item representation are sent to matrix factorization and neural network for recommendation. The experimental results on the public dataset MovieLens show that the algorithm can effectively improve the accuracy of the recommendation results.