• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学 ›› 2020, Vol. 42 ›› Issue (08): 1500-1505.

• 人工智能与数据挖掘 • 上一篇    下一篇

一种基于降噪自编码器的隐式反馈推荐算法

杨丰瑞1,2,3,李前洋1,2,罗思烦1,2   

  1. (1.重庆邮电大学通信与信息工程学院,重庆 400065;2.重庆邮电大学通信新技术应用研究中心,重庆 400065;

    3.重庆重邮信科(集团)股份有限公司,重庆 401121)
  • 收稿日期:2019-12-31 修回日期:2020-03-25 接受日期:2020-08-25 出版日期:2020-08-25 发布日期:2020-08-29

An implicit feedback recommendation algorithm based on denoising autoencoder

YANG Feng-rui1,2,3,LI Qian-yang1,2,LUO Si-fan1,2   

  1. (1.School of Communication and Information Engineering,
    Chongqing University of Posts and Telecommunications,Chongqing 400065;
    2.Research Center of New Telecommunication Technology,
    Chongqing University of Posts and Telecommunications,Chongqing 400065;
    3.Chongqing University of Posts and Telecommunications Information Technology(Group)Co.,Ltd.,Chongqing 401121,China)

  • Received:2019-12-31 Revised:2020-03-25 Accepted:2020-08-25 Online:2020-08-25 Published:2020-08-29

摘要: 现有的隐式反馈协同算法直接利用稀疏的二值社交信任信息辅助推荐,存在严重的数据稀疏问题,且没有深层次地融合社交信任信息的影响。针对以上问题,提出利用降噪自编码器深度融合用户隐式反馈数据与社交信息的算法。首先从不同的角度区分用户信任,提出一种信任相似度的新度量方法来改善社交数据的稀疏性,利用降噪自编码器将信任数据与用户隐式交互信息深度融合,通过综合二者的影响,有效提高了推荐质量。实验表明,该算法优于现有主流的的隐式反馈推荐算法。


关键词: 推荐系统, 社交网络, 深度学习, 降噪自编码器, 隐式反馈

Abstract: The existing implicit feedback collaborative algorithms directly use sparse binary social trust information to assist in recommendation, which has serious data sparseness problems. Besides, there is no impact of deep integration of social trust information. Aiming at the above problems, an algorithm is proposed, which uses the denoising autoencoder to deeply fuse the user's implicit feedback data and social information. Firstly, we distinguish the trust of users from different angles, propose a new measurement method of trust similarity, and improve the sparseness of social trust data. Secondly, denoising autoencoders are used to deeply fuse trust data and user implicit interactive information. By combining the advantages of the two, the recommendation quality is effectively improved. Experiments show that this method is superior to the existing mainstream implicit feedback recommendation algorithms.


Key words: recommendation system, social network, deep learning, denoising autoencoder, implicit feedback