ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2016, Vol. 53 ›› Issue (6): 1389-1399.doi: 10.7544/issn1000-1239.2016.20150307

• 信息处理 • 上一篇    下一篇

一种基于多元社交信任的协同过滤推荐算法

王瑞琴1,蒋云良1,李一啸2,楼俊钢1   

  1. 1(湖州师范学院信息工程学院 浙江湖州 313000);2(浙江财经大学信息管理与工程学院 杭州 310018) (angelwrq@163.com)
  • 出版日期: 2016-06-01
  • 基金资助: 
    国家自然科学基金项目(61402336,61370173,61403338);国家教育部科学基金项目(14YJCZH152);浙江省自然科学基金项目(LY15F020018);浙江省科技计划项目(2013C31138,2015C33247)

A Collaborative Filtering Recommendation Algorithm Based on Multiple Social Trusts

Wang Ruiqin1, Jiang Yunliang1, Li Yixiao2, Lou Jungang1   

  1. 1(School of Information Engineering, Huzhou University, Huzhou, Zhejiang 313000);2(School of Information Management and Engineering, Zhejiang University of Finance & Economics, Hangzhou 310018)
  • Online: 2016-06-01

摘要: 协同过滤推荐是当前最成功的个性化推荐技术之一,但是传统的协同过滤推荐算法普遍存在推荐性能低和抗攻击能力弱的问题.针对以上问题,提出了一种基于多元化社交信任的协同过滤推荐算法CF-CRIS (collaborative filtering based on credibility, reliability, intimacy and self-orientation).1)借鉴社会心理学中的信任产生原理,提出基于多个信任要素(可信度、可靠度、亲密度、自我意识导向)的信任度计算方法;2)深入研究社交网络环境中各信任要素的识别、提取和量化方法;3)基于用户间的综合信任度选取可信邻居,完成对目标用户的个性化推荐.基于通用测试数据集的实验研究结果表明:该算法不但可以极大地提高推荐系统的精确度和召回率,而且表现出良好的抗攻击能力.

关键词: 协同过滤, 社交网络, 信任, 信任要素, 推荐精度, 召回率, 抗攻击能力

Abstract: Collaborative filtering (CF) is one of the most successful recommendation technologies in the personalized recommendation systems. It can recommend products or information for target user according to the preference information of similar users. However the traditional collaborative filtering algorithms have the disadvantages of low recommendation efficiency and weak capacity of attack-resistance. In order to solve the above problems, a novel collaborative filtering algorithm based on social trusts is proposed. Firstly, referring to the trust generation principle in social psychology, a social trust computation method based on multiple trust elements is presented. In social networking environment, trust elements mainly include credibility, reliability, intimacy and self-orientation. Then specific methods of identifying, extraction and quantification of the trust elements are studied in depth. Finally, the trustworthy neighbors of target user are selected in accordance with the social trust, so as to make trust-based collaborative recommendation. Using the FilmTrust and Epinions as test data sets, the performance of the novel algorithm is compared with that of the traditional CF and the-state-of-art methods, as well as the CF based on single trust element. Experimental results show that compared with the other methods, the proposed algorithm not only improves the recommendation precision and recall, but also has powerful attack-resistance capacity.

Key words: collaborative filtering (CF), social network, trust, trust elements, recommendation precision, recall, attack-resistance capacity

中图分类号: