电子学报

• 学术论文 • 上一篇    下一篇

基于受限信任关系和概率分解矩阵的推荐

印桂生, 张亚楠, 董宇欣, 韩启龙   

  1. 哈尔滨工程大学计算机科学与技术学院,黑龙江哈尔滨 150001
  • 收稿日期:2013-05-31 修回日期:2013-07-17 出版日期:2014-05-25
    • 作者简介:
    • 印桂生 男,1964年生于江苏省泰兴市.哈尔滨工程大学计算机科学与技术学院,教授、博士生导师.研究方向为数据库与知识发现,虚拟现实,网购软件.张亚楠(通信作者) 男,1981年生于黑龙江省哈尔滨市.哈尔滨工程大学计算机科学与技术学院,博士生.研究方向为数据挖掘、推荐系统. E-mail:ynzhang-1981@163.com
    • 基金资助:
    • 国家自然科学基金 (No.612721836,No.61100007); 黑龙江省博士后基金 (No.LBH-Z12068); 哈尔滨工程大学自由探索基金 (No.HEUCF100608)

A Constrained Trust Recommendation Using Probabilistic Matrix Factorization

YIN Gui-sheng, ZHANG Ya-nan, DONG Yu-xin, HAN Qi-long   

  1. College of computer Science and Technology, Harbin Engineering University, Harbin, Heilongjiang 150001, China
  • Received:2013-05-31 Revised:2013-07-17 Online:2014-05-25 Published:2014-05-25
    • Supported by:
    • National Natural Science Foundation of China (No.612721836, No.61100007); Post-doctoral Foundation of Heilongjiang Province (No.LBH-Z12068); Free Exploration Foundation of Harbin Engineering University (No.HEUCF100608)

摘要: 现有的推荐算法很难对没有任何记录的冷启动用户或者历史记录稀疏的用户给出准确的推荐,即用户的冷启动问题.本文提出一种基于受限信任关系和概率分解矩阵的推荐方法,由不信任关系约束信任关系的传播,得到准确且覆盖全面的用户信任关系矩阵,并通过对用户信任关系矩阵和用户商品矩阵的概率分解联合用户信任关系和用户商品矩阵信息,为用户给出推荐.实验表明该方法对冷启动用户和历史记录稀疏的用户的推荐效果有较大幅度的提升,有效地解决了用户的冷启动问题.

关键词: 推荐算法, 受限信任传播, 概率分解矩阵, 用户的冷启动问题

Abstract: Existing recommendation algorithms can not give accurate recommendations for users who have few historical records or even none,namely the user cold recommendation problem.In this paper,a constrained trust recommendation using probabilistic matrix factorization (CTRPMF) is proposed.The trust is propagated with the constraint of distrust to get accurate and comprehensive trust relationship matrix.User trust relationship matrix and user-item matrix are factorized using probabilistic matrix factorization to mix the information from trust relationship and user-item matrix,in order to give recommendations.The experimental results showed that CTRPMF could greatly improve the effectiveness of recommend ations for cold start users and users with sparse historical data,and effectively solve the cold recommendation problem.

Key words: recommendation algorithm, constrained trust propagation, probabilistic matrix factorization, user cold start problem

中图分类号: