基于联合概率矩阵分解的移动社会化推荐  

Mobile Social Recommendation Based on Unified Probabilistic Matrix Factorization

作　　者：熊丽荣[1] 刘坚[1] 汤颖[1] 

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：《计算机科学》2016年第9期255-260,265,共7页Computer Science

基　　金：浙江省重大科技专项重大工业项目(2012C11026-2)资助

摘　　要：利用移动设备上下文、移动社会化网络等信息进一步提高推荐系统的预测准确率,并缓解可能存在的数据稀疏性和冷启动问题,已经成为移动推荐系统的主要任务。采用基于矩阵分解的因子分析方法,结合用户、服务和用户社会化网络信息进行服务推荐,可以缓解数据稀疏性和冷启动问题;同时,为了增加信任矩阵密度,引入间接信任关系,提出了一种符合移动社会化网络特点的信任度计算方法,该方法仅利用移动社会化网络结构信息构建信任矩阵,从而减少用户对信任关系的主动标识。实验结果表明,引入间接信任关系能够提高预测精度,同时比传统的协同过滤算法和已有的一些矩阵分解方法具有更好的预测准确率,特别是在评分数据稀疏的情况下。It has become the main task of mobile recommender systems to further improve the prediction quality and solve the data sparsity and cold-start problems that may exist by employing mobile context and mobile social network information etc. We combined users, services and users＇ social network information for recommendation to alleviate the data sparsity and cold-start problems by using the factor analysis method based on matrix factorization （MF）. In order to increase the trust matrix density,in this paper we imported the indirect trust relationship,and then proposed a trust relationship calculation method which only use the mobile social network information to build trust matrix to reduce the user＇s active identification for trust relationship. And the trust calculation method is in line with the characteristics of mobile social network. The experimental results show that the introduction of the indirect trust relationship can improve the prediction accuracy, and our method outperforms some existing MF methods and traditional collaborative filtering algorithm in the aspect of accuracy, especially in the circumstance that users have made very few ratings or even none at all.

关 键 词：移动推荐 社会化推荐 矩阵分解 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]