基于用户影响力和偏好一致性的社会化推荐  

Social Recommendation Based on User Influence and Preference Consistency

作　　者：孙晶晶 荀亚玲[1] 杨海峰[1] SUN Jing-jing;XUN Ya-ling;YANG Hai-feng(School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学计算机科学与技术学院,山西太原030024

出　　处：《计算机技术与发展》2023年第9期91-97,共7页Computer Technology and Development

基　　金：国家自然科学基金项目(62272336);山西省自然科学基金(201901D211302)。

摘　　要：用户和项目的急剧增加使得评分数据过于稀疏导致传统推荐算法效果较差,社交网络信息的引入缓解了传统推荐系统中面临的数据稀疏性问题。然而,现有社会化推荐在刻画用户之间的信任关系时未考虑到用户之间的信任具有偏好差异性和信任传播稳定性不强等问题。因此,提出一种基于用户影响力和偏好一致性的社会化推荐。首先,结合评分信息和社交信息从偏好一致性方向刻画用户之间的信任强度,挖掘出隐藏的信息,缓解了用户的偏好差异性。其次,借助用户的社会影响力找到一条信任传播稳定性最强的路径,避免信任在传播过程中造成信任节点信息的丢失。然后,将用户的评分相似度和信任相似度线性加权得到用户的近邻用户做评分预测。最后,将该方法与现有社会化推荐算法在Filmtrust和CiaoDVD数据集上进行综合实验,结果表明该方法在MAE和RMSE上优于现有推荐算法。The sharp increase of users and items makes the rating data too sparse,which leads to poor performance of traditional recommendation algorithms.The introduction of social network information alleviates the data sparsity problem faced by traditional recommendation systems.However,the existing social recommendation does not take into account the differences in preferences between users and the weak stability of trust propagation when describing the trust relationship between users.Therefore,a social recommendation based on user influence and preference consistency is proposed.Firstly,combining the rating information and social information,the trust strength between users is described from the direction of preference consistency,and the hidden information is mined to alleviate the user's preference difference.Secondly,with the social influence of users,a path with the strongest trust propagation stability is found,which avoids the loss of trust node information in the process of trust propagation.Then,the user’s rating similarity and trust similarity are linearly weighted to obtain the user’s neighbors for rating prediction.Finally,the proposed method and the existing social recommendation algorithm are comprehensively tested on the Filmtrust and CiaoDVD datasets.It is showed that the proposed method outperforms the existing recommendation algorithms in MAE and RMSE.

关 键 词：社会化推荐 综合信任 协同过滤 偏好一致性 用户影响力 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]