融合用户特征与知识图谱的可解释电影推荐方法  

Explainable movie recommendation method combining user features and knowledge graph

作　　者：罗震 张涛[1,2] LUO Zhen;ZHANG Tao(School of Information Management and Engineering,Shanghai University of Finance and Economics,Shanghai 200433,China;Shanghai Key Laboratory of Financial Information Technology,Shanghai University of Finance and Economics,Shanghai 200433,China)

机构地区：[1]上海财经大学信息管理与工程学院,上海200433 [2]上海财经大学上海市金融信息技术研究重点实验室,上海200433

出　　处：《管理工程学报》2024年第6期128-139,共12页Journal of Industrial Engineering and Engineering Management

基　　金：上海市自然科学基金项目(19ZR1417200);教育部人文社会科学研究规划基金项目(19YJA630116);中央高校基本科研业务费专项资金项目(2023110139)。

摘　　要：可解释的电影推荐系统可以帮助用户过滤无关信息并给出推荐理由,与用户建立信任。基于知识图谱的推荐方法能充分利用电影信息的公开性和丰富性,但现有研究通常忽略用户层面的信息,解释机制对用户也不友好。针对目前存在的问题,本文引入了用户的交叉特征和基于电影特征构造的知识图谱,提出了一个新的可解释的电影推荐方法——EKPCF(enhanced knowledge path with cross feature),结合知识图谱嵌入和语义路径挖掘用户的潜在偏好,利用高贡献的交叉特征和语义路径解释推荐结果。该方法与其他相关研究工作的不同之处在于以下几个方面:(1)为了提高准确率,改进路径筛选机制,将用户层面的交叉特征与知识图谱语义路径相结合,更新用户表征;(2)为了提高可解释性,利用数据挖掘方法得到新的用户特征,并将知识图谱语义路径中的陌生节点剔除;(3)将推荐机制透明化,借助交叉特征与高分语义路径共同解释推荐结果。最后,本文使用两个真实的数据集验证了EKPCF方法的有效性,同时展示该方法的解释机制。The abundant network resources may often lead to“information overload”,resulting in users not being able to quickly identify their needs,so the recommendation system emerges as required.In the online market,the competition is fierce,and service providers need to capture users′psychology and improve their satisfaction and trust in order to obtain more user resources.However,it is never sufficient to solely rely on personalized recommendations.The transparency of recommendation mechanism and the explainability of recommendation results should also be taken into account.Explainable movie recommendation system can help users filter irrelevant information and give reasons for recommendation,which help build trust with users.Knowledge graph-based recommendation methods can make full use of the openness and richness of movie information,but existing studies usually ignore user-level information and the explanation mechanism is not user-friendly.To address the current problems,this paper introduces cross features of users and knowledge graphs constructed based on movie features,and proposes a new explainable movie recommendation method-EKPCF(Enhanced Knowledge Path with Cross Feature).This method combines knowledge graph embedding and semantic paths to mine users′potential preferences,and uses highly contributed cross features and semantic paths to explain recommendation results.Such approach differs from other related research works in the following aspects:1)to enhance the accuracy,the path selection mechanism is improved by combining the user-level cross features with the semantic path of the knowledge graph to update the user representation;2)to improve the explainability,new user features are obtained using data mining methods and unfamiliar nodes in the semantic path of the knowledge graph are removed;3)the recommendation mechanism is made transparent,and the recommendation results are explained by means of cross features and high scoring semantic paths together.The first part introduces the background and

关 键 词：推荐系统 可解释性 知识图谱 注意力机制 循环神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] C93[自动化与计算机技术—计算机科学与技术]