一种融合用户偏好和信任-不信任关系的社会化推荐方法  

Social Recommendation Method Combining User Preference and Trust-distrust Relationship

作　　者：虞胜杰 熊丽荣[1] 王玲燕 YU Sheng-jie;XIONG Li-rong;WANG Ling-yan(College of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]浙江工业大学计算机科学与技术学院,杭州310023

出　　处：《小型微型计算机系统》2020年第12期2529-2535,共7页Journal of Chinese Computer Systems

摘　　要：引入辅助数据如社交关系信息是解决传统个性化推荐算法数据稀疏问题的一种有效方法.目前大部分基于信任的推荐算法直接利用二值信任网络来提升推荐质量,然而,直接信任关系也是稀疏的.因此,本文利用社交网络中的信任信息,挖掘用户在推荐系统中的隐式信任关系,增加信任数据的密度.考虑到用户偏好信息的稀疏性,本文利用社交网络中信任信息和评分信息,结合矩阵分解的特征因子分析法缓解该影响.在线网络也存在不信任关系,如电子商务网站Epinions等.不信任信息对社交网络有重要意义,但很少有算法将不信任关系引入到推荐中.本文利用不信任信息对信任网络进行调节,选择符合社会平衡理论的稳定的路径,利用Jaccard系数的变体计算用户间的不信任权重,对于不信任权重大于设定阈值的用户,否定其局部信任度量的结果.实验结论证明,本文的信任模型可以提高预测精度,相比于融合信任的传统协同过滤算法和已有的信任增强的矩阵分解推荐方法,本文的算法具有更好的推荐准确率.The introduction of auxiliary data such as social relationship information is an effective method to solve data sparse problem of traditional personalized recommendation algorithms.M ost trust-based recommendation algorithms directly use binary trust network information to improve the recommendation quality.However,direct trust relationship is also sparse.Therefore,this paper mines users’implicit trust relationships to increase the density of trust data.Considering the sparseness of user preference information,this paper uses matrix factorization(M F)method combined with the trust information and score information in social networks to alleviate this problem.There are also distrust relationships in online network,such as e-commerce site Epinions.Distrust information is important for social network,but fewalgorithms introduce distrust relationships into recommendation.Therefore,this paper utilizes distrust information to debug the trust network.This paper selects all stable paths that conform to the social balance theory,and then uses the variant of the Jaccard coefficient to calculate the distrust weights between users.For users w hose distrust weights are larger than the set threshold,the results of local trust metrics are negated.Experimental results show that this paper’s trust model can improve the prediction accuracy,and this paper’s algorithm outperforms some existing M F methods with trust enhancement and traditional collaborative filtering algorithm.

关 键 词：社会化推荐 矩阵分解 信任度 不信任度 数据稀疏性 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]