基于知识图谱和协同过滤的电影推荐算法研究

A movie recommendation algorithm based on knowledge graph and collaborative filtering

针对协同过滤算法在推荐电影过程中只能考虑电影外部评论而不能考虑电影内部的相似度关系,提出构建知识图谱辅助计算电影内部相似度。已有的电影数据可能是不完整的,因此采用知识图谱推理补全缺失的电影知识。基于TransE模型的知识图谱无法有效描述电影间的片名、演员、导演等复杂的多关系。首先采用改进的TransHR模型表示出电影信息之间的多关系,提升关系表示的准确率;然后通过用户评分矩阵计算电影间相似度;最后将2种相似度融合并应用于矩阵分解的推荐技术中。对比实验结果表明,该算法在召回率、准确率、平均绝对误差MAE等指标上都有所提升。

Aiming at the problem that the collaborative filtering recommendation algorithm can only consider the external reviews of movies and not the similarity relationships within movies during the process of recommending movies, the paper proposes to construct a knowledge graph to help calculate the internal similarity of movies. Existing movie data may be incomplete, so knowledge graph reasoning is used to complement missing movie knowledge. The knowledge graph based on TransE model cannot effectively describe the complex multi-relationship among movie titles, actors and directors. Firstly, the improved TransHR model can express the multi-relationship between movie information and improve the accuracy of relationship representation. Then, the similarity between movies is calculated by the user rating matrix. Finally, the two similarities are merged and applied to the recommended technique of matrix decomposition. The experimental results show that the algorithm improves Recall, Precision, and MAE.