融合评论文本和社交网络的矩阵分解推荐方法  

Matrix Factorization Recommendation Method by Fusing Review Text and Social Networks

作　　者：马宁 付伟[1] 季伟东[1] 丁云鸿 朱海龙 严武尉 李超 杨耀 MA Ning;FU Wei;JI Wei-dong;DING Yun-hong;ZHU Hai-long;YAN Wu-wei;LI Chao;YANG Yao(School of Computer Science and Information Engineering,Harbin Normal University,Harbin 150025,China)

机构地区：[1]哈尔滨师范大学计算机科学与信息工程学院,哈尔滨150025

出　　处：《小型微型计算机系统》2022年第2期312-319,共8页Journal of Chinese Computer Systems

基　　金：黑龙江省自然科学基金项目(F2018002)资助;黑龙江省高等教育教学改革项目(SJGY20180259)资助;哈尔滨市科技局科技创新人才研究专项项目(2017RAQXJ050)资助;哈尔滨师范大学博士科研启动基金项目(XKB201901)资助;哈尔滨师范大学计算机学院科研项目(JKYKYY202006)资助;哈尔滨师范大学研究生培养质量提升工程项目(HSDYJSJG2019006)资助。

摘　　要：基于矩阵分解的推荐方法易受到数据稀疏性问题的影响,常见的解决办法是向矩阵分解模型中融入评论文本信息,但是这类方法通常假设用户是独立存在的,忽略了用户之间的社交关系.现实世界中用户的行为与喜好往往会受到其信任好友的影响,因此本文提出一种融合评论文本和社交网络的矩阵分解推荐方法(Review and social probabilistic matrix factorization,RSPMF).首先设计了深度神经网络模型用于学习评论文本的上下文特征;其次,设计了信任传播模型用于根据社交好友的特征修正用户的潜在隐特征;最后将上述两种模型以正则化方式融入概率矩阵分解模型,通过训练模型获取用户与物品之间的内在关系并实现物品推荐.在公开的真实数据集Yelp上进行了实验,并与多种前沿的算法进行了性能对比,结果表明本文提出的RSPMF方法具有良好的推荐性能.The recommendation method based on matrix factorization is easily affected by data sparsity.A common solution is to incorporate review text information into the matrix factorization model.However,such methods usually assume that users exist independently,ignoring the social relationship between users.In the real world,users′behaviors and preferences are often affected by their trusted friends.Therefore,this paper proposes a matrix factorization recommendation method which integrates reviews and social networks.Firstly,a deep neural networks model is designed to learn the context features of reviews.Secondly,a trust propagation model is designed to modify the latent features of users according to the characteristics of social friends.Finally,the above two models are incorporated into the probability matrix factorization model by regularization,and the internal relationship between users and items is obtained by training the model,and the item recommendation is realized.Experiments are carried out on the real-world dataset Yelp,and the results show that the proposed RSPMF method has better recommendation performance compared with other advanced algorithms.

关 键 词：推荐系统 深度神经网络 矩阵分解 社交网络 信任传播 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]