融合隐式信任与属性偏好的群组推荐算法  

Group recommendation algorithm combining implicit trust and attribute preference

作　　者：边纪超 庞继芳[1,2] 宋鹏[3] Bian Jichao;Pang Jifang;Song Peng(School of Computer and Information Technology,Shanxi University,Taiyuan,030006,China;Key Laboratory of Computational Intelligence and Chinese Information Processing of Ministry of Education,Shanxi University,Taiyuan,030006,China;School of Economics and Management,Shanxi University,Taiyuan,030006,China)

机构地区：[1]山西大学计算机与信息技术学院,太原030006 [2]山西大学计算智能与中文信息处理教育部重点实验室,太原030006 [3]山西大学经济与管理学院,太原030006

出　　处：《南京大学学报（自然科学版）》2023年第5期803-812,共10页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金(62006148,72171137);山西省1331工程。

摘　　要：随着互联网和推荐系统的不断发展,推荐服务的对象由单一用户扩展为群组成员,获取并融合组内成员的偏好、提升群组推荐效果成为当前推荐领域研究的热点问题.利用用户提供的多属性评分矩阵,提出一种融合隐式信任与属性偏好的群组推荐算法.首先,基于用户共同评分项目数和多属性评分相似度计算用户间的直接隐式信任,并利用信任传递机制获取用户间的间接信任,降低数据稀疏性.然后,通过计算用户各属性评分与总体评分间的距离来挖掘用户的属性偏好,在此基础上,利用注意力机制学习组内用户权重,将用户偏好聚合为群组偏好,进而结合深度学习框架对候选项目进行预测,生成最终的推荐列表.最后,四个数据集上的实验验证了提出的算法的有效性和可行性,实验结果表明,该算法的准确率、nDCG等评价指标明显优于对比算法.With the continuous development of the Internet and recommendation system,the object of recommendation service expands from a single user to group members.How to obtain and integrate the preferences of group members and improve the effect of group recommendation has become a hot issue in the field of recommendation research.This paper makes full use of the user-provided multi-attribute rating matrix,and a group recommendation algorithm combining implicit trust and attribute preference is proposed.Firstly,the direct implicit trust between users is calculated based on the number of items shared by users and the similarity of multi-attribute ratings.In order to reduce the sparsity of data,the trust transfer mechanism is used to obtain indirect trust between users.Then,the user's attribute preference is mined by calculating the distance between each attribute rating and the overall rating.On this basis,the attention mechanism is used to learn the weight of users in the group,and user preferences are aggregated into group preferences.Then the deep learning framework is combined to predict candidate projects and generate the final recommendation list.Finally,experiments conducted on four datasets to verify the effectiveness and feasibility of the proposed algorithm verify its significantly superior to the compared algorithm in accuracy,nDCG and other evaluation indicators.

关 键 词：群组推荐 多属性评分矩阵 隐式信任 属性偏好 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]