基于知识图谱的推荐算法研究  

Research of Recommendation Algorithm Based on Knowledge Graph

作　　者：文峰[1] 曹雄 黄海新[1] 闫啸云 WEN Feng;CAO Xiong;HUANG Haixin;YAN Xiaoyun(Shenyang Ligong University,Shenyang 110159,China;China University of Petroleum,Beijing 102249,China)

机构地区：[1]沈阳理工大学信息科学与工程学院,沈阳110159 [2]中国石油大学信息科学与工程学院,北京102249

出　　处：《沈阳理工大学学报》2021年第6期13-17,共5页Journal of Shenyang Ligong University

基　　金：辽宁省高等学校创新人才支持计划;沈阳理工大学科研创新团队建设计划资助项目(SYLUTD202105)。

摘　　要：随着互联网的发展,数据开始呈爆炸性的方式增长,随之而来的是海量的脏数据。为了不受这些脏数据的影响得到高质量的数据,推荐系统开始引起人们的广泛关注。传统的推荐算法比如协同过滤(CF)太过于依赖用户和物品的交互信息,因此遗留的数据稀疏和冷启动问题成为学术界一直需要攻克的难题。最近,知识图谱由于其三元组的易理解性以及丰富的语义信息而被广泛应用在推荐系统中。由此提出一种基于知识图谱嵌入模型TransE和图神经网络的推荐算法,使用图神经网络可以提取用户和物品的高阶特征,知识图谱嵌入算法模型可以使提取出的特征信息更加丰富,在相关数据集上的实验结果表明该算法的推荐性能表现优异。With the development of the Internet,data began to grow along with huse amounts of dirty data in an explosive way.In order to get higher quality data without being affected by these dirty data,people begin to focuson Recommendation Systems.Traditional recommendation algorithms,such as Collaborative Filtering(CF),rely too much on the interactive information between users and items,so the problems of data sparsity and cold start have become difficult problems to be solved by the academia all the time.Recently,Knowledge Graph has been widely used in recommendation systems because of its trituple′s intelligibility and rich semantic information.For these reasons,we propose a recommendation algorithm model,which is based on knowledge embedding model TransE and Graph neural network.Higher-order representation of users and items can be extracted by using the Graph neural network.Knowledge graph embedding algorithm model can extract the characteristic information in more abundant way.Experimental results on related datasets show that the recommended performance of the algorithm is excellent.

关 键 词：推荐系统 知识图谱 图神经网络 图嵌入 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]