摘要
现有的推荐算法仅依靠历史行为无法挖掘到用户的深层兴趣,且受到用户冷启动问题的制约,针对上述问题提出融合知识图谱的用户多层兴趣模型(MIKU).该模型首先以用户的历史交互项目为知识图谱的头实体构建用户浅层兴趣,结合知识图谱中关系路径链接到历史项目的相关实体挖掘用户深层兴趣;其次考虑到用户兴趣的多样性,针对不同层次的兴趣分别采用自适应加权机制,学习用户对每个历史行为以及各个深层兴趣点的关注度.该模型在细粒度刻画物品特征的同时,利用知识图谱的结构信息为用户的推荐结果提供了一定的可解释性,并且结合了用户的属性特征有效解决了冷启动问题.通过在公开的MovieLens-1M数据集上进行验证,结果表明MIKU模型同CKE、RippleNet等基准模型相比,在推荐结果的准确率上提高了1.93%~5.59%、召回率上提升了2.95%~4.7%.
Existing recommendation algorithms cannot dig out the implicit interest of users when only relying on historical behavior.At the same time,it will be restricted by user′s cold start problems.To solve these problems,this paper proposes a multi-layer user interest model based on knowledge graph(MIKU).Firstly,this model takes the user′s historical interactive items as the head entity of the knowledge graph to construct direct interest,and uses the relational path in the knowledge graph to link to the related entities of the historical item to construct deep interests.Secondly,considering the diversity of user interests we use adaptive weighting mechanisms for different levels of interest in order that learn the user′s preference for each historical behavior and calculate the user′s attention to deep-level points of interest.The algorithm describes the characteristics of items in a fine-grained manner,and uses the structural information of the knowledge graph to provide a certain interpretability for the user′s recommendation results.Mean-while,it combines the user′s attribute characteristics to solve the cold start problem effectively.Through extensive experiments on a public MovieLens-1M dataset,we demonstrate that compared with several state-of-the-art baselines,such as CKE and RippleNet,the MIKU model has an improvement of1.93%~5.59%in the accuracy,and the recall rate has increased by2.95%~4.7%.
作者
段文菁
谢珺
续欣莹
岳晓冬
刘笑笑
DUAN Wen-jing;XIE Jun;XU Xin-ying;YUE Xiao-dong;LIU Xiao-xiao(College of Information and Computer,Taiyuan University of Technology,Jinzhong 030600,China;College of Electrical and Power Engineering,Taiyuan University of Technology,Taiyuan 030000,China;School of Computer Engineering and Science,Shanghai University,Shanghai 200444,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2022年第5期1006-1012,共7页
Journal of Chinese Computer Systems
基金
山西省应用基础研究项目(201801D221190,201801D121144)资助。
关键词
知识图谱
用户属性
深层兴趣
自适应权重
knowledge graph
user attributes
deep interest
adaptive weigh
