Computer Engineering and Applications ›› 2020, Vol. 56 ›› Issue (2): 106-114.DOI: 10.3778/j.issn.1002-8331.1907-0131

Previous Articles     Next Articles

Fusion Recurrent Knowledge Graph and Collaborative Filtering Movie Recommendation Algorithm

LI Hao, ZHANG Yachuan, KANG Yan, YANG Bing, BU Rongjing, LI Jinyuan   

  1. School of Software, Yunnan University, Kunming 650091, China
  • Online:2020-01-15 Published:2020-01-14



  1. 云南大学 软件学院,昆明 650091

Abstract: Recommendation system has significance in screening the useful information and improving the efficiency of information acquisition. However, sparse data and cold start are encountered in traditional recommendation systems. Therefore, combining external rating and item connotation knowledge, this paper proposes a movie recommendation model based on recurrent knowledge graph and collaborative filtering——RKGE-CF. After the full consideration on correlation between items, users and ratings, Top-[K] recommendation is adopted which using collaborative filtering based on items and users. It adds the items external additional data and user preference data to the knowledge map, extracts the dependencies between entities and builds the interactive information between users and items. Then, the model can reveal the semantics between entities and relations, improve the understanding of users’ interests to make recommendations. Different algorithms are used to fuse several recommendation results and compare it. In the training of the model, it uses multiple groups of different negative samples for comparison, to optimize the model. Finally, a new dataset is obtained for testing by mapping real movie Movielens and IMDB. Experimental results show that this model improves the accuracy of recommendation effect, and explains the reasons behind the recommendation.

Key words: knowledge graph, collaborative filtering, recommendation system, explainable recommendation

摘要: 推荐系统对筛选有效信息和提高信息获取效率具有重大的意义。传统的推荐系统会面临数据稀松和冷启动等问题。利用外部评分和物品内涵知识相结合,提出一种基于循环知识图谱和协同过滤的电影推荐模型——RKGE-CF。在充分考虑物品、用户、评分之间的相关性后,利用基于物品和用户的协同过滤进行Top-[K]推荐;将物品的外部附加数据和用户偏好数据加入知识图谱,提取实体相互之间的依赖关系,构建用户和物品之间的交互信息,以便揭示实体与关系之间的语义,帮助理解用户兴趣;将多种推荐结果按不同方法融合进行对比;模型训练时使用多组不同的负样本作为对比,以优化模型;最后利用真实电影数Movielens和IMDB映射连接成新数据集进行测试。实验结果证明该模型对于推荐效果的准确率有显著的提升,同时能更好地解释推荐背后的原因。

关键词: 知识图谱, 协同过滤, 推荐系统, 可解释性推荐