基于知识表示学习的协同矩阵分解方法  

Collective Matrix Factorization Based on Knowledge Representation Learning

作　　者：刘琼昕 覃明帅[2,3] LIU Qiongxin;QIN Mingshuai(Beijing Engineering Research Center of High Volume Language Information Processing and Cloud Computing Applications,Beijing 100081,China;School of Computer Science and Technology,Beijing Institute of Technology,Beijing 100081,China;The Key Laboratory of Rich-Media Knowledge Organization and Service of Digital Publishing Content,Institute of Scientific&Technical Information of China,Beijing 100038,China)

机构地区：[1]北京市海量语言信息处理与云计算应用工程技术研究中心,北京100081 [2]北京理工大学计算机学院,北京100081 [3]中国科学技术信息研究所富媒体数字出版内容组织与知识服务重点实验室,北京100038

出　　处：《北京理工大学学报》2021年第7期752-757,共6页Transactions of Beijing Institute of Technology

基　　金：国家部委预研项目(3151109020)。

摘　　要：针对协同过滤算法中用户反馈数据的稀疏性问题,提出一种基于知识库的协同矩阵分解方法.该方法从物品的知识图谱中学习其向量表示,并在此基础上联合地分解反馈矩阵和物品关联度矩阵,两种矩阵共享物品向量,利用物品的语义信息弥补反馈数据的缺失.实验结果表明,该方法显著地提升了矩阵分解模型的推荐效果,在一定程度上解决了协同过滤的冷启动问题.In order to solve the problem of user feedback data sparseness existed in collaborative filtering method,a collective matrix factorization method was proposed based on knowledge graph.The method was arranged to make up for the scarce of the user feedback data with additional item sematic information.Learning item embeddings from items’knowledge graph,the method was designed to jointly factorize a user feedback matrix and an item relatedness matrix with the same item embeddings.Experimental results on two datasets show that the proposed method can significantly improve the performance of matrix factorization models,and it can solve the cold start problem to some extent.

关 键 词：推荐系统 矩阵分解 知识表示学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]