Dual Metric Learning for Effective and Efficient Cross-Domain Recommendations

被引:26
作者
Li, Pan [1 ]
Tuzhilin, Alexander [1 ]
机构
[1] NYU, Stern Sch Business, Dept Technol Operat & Stat, New York, NY 10012 USA
基金
澳大利亚研究理事会;
关键词
Cross domain recommendation; dual learning; metric learning; orthogonal mapping; ALGORITHMS;
D O I
10.1109/TKDE.2021.3074395
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Cross domain recommender systems have been increasingly valuable for helping consumers identify useful items in different applications. However, existing cross-domain models typically require large number of overlap users, which can be difficult to obtain in some applications. In addition, they did not consider the duality structure of cross-domain recommendation tasks, thus failing to take into account bidirectional latent relations between users and items and achieve optimal recommendation performance. To address these issues, in this paper we propose a novel cross-domain recommendation model based on dual learning that transfers information between two related domains in an iterative manner until the learning process stabilizes. We develop a novel latent orthogonal mapping to extract user preferences over multiple domains while preserving relations between users across different latent spaces. Furthermore, we combine the dual learning method with the metric learning approach, which allows us to significantly reduce the required common user overlap across the two domains and leads to even better cross-domain recommendation performance. We test the proposed model on two large-scale industrial datasets and six domain pairs, demonstrating that it consistently and significantly outperforms all the state-of-the-art baselines. We also show that the proposed model works well with very few overlap users to obtain satisfying recommendation performance comparable to the state-of-the-art baselines that use many overlap users.
引用
收藏
页码:321 / 334
页数:14
相关论文
共 78 条
  • [71] Graph Convolutional Neural Networks for Web-Scale Recommender Systems
    Ying, Rex
    He, Ruining
    Chen, Kaifeng
    Eksombatchai, Pong
    Hamilton, William L.
    Leskovec, Jure
    [J]. KDD'18: PROCEEDINGS OF THE 24TH ACM SIGKDD INTERNATIONAL CONFERENCE ON KNOWLEDGE DISCOVERY & DATA MINING, 2018, : 974 - 983
  • [72] Deep Learning Based Recommender System: A Survey and New Perspectives
    Zhang, Shuai
    Yao, Lina
    Sun, Aixin
    Tay, Yi
    [J]. ACM COMPUTING SURVEYS, 2019, 52 (01)
  • [73] CATN: Cross-Domain Recommendation for Cold-Start Users via Aspect Transfer Network
    Zhao, Cheng
    Li, Chenliang
    Xiao, Rong
    Deng, Hongbo
    Sun, Aixin
    [J]. PROCEEDINGS OF THE 43RD INTERNATIONAL ACM SIGIR CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION RETRIEVAL (SIGIR '20), 2020, : 229 - 238
  • [74] Cross-Domain Recommendation via Preference Propagation GraphNet
    Zhao, Cheng
    Li, Chenliang
    Fu, Cong
    [J]. PROCEEDINGS OF THE 28TH ACM INTERNATIONAL CONFERENCE ON INFORMATION & KNOWLEDGE MANAGEMENT (CIKM '19), 2019, : 2165 - 2168
  • [75] Zhong EH, 2009, KDD-09: 15TH ACM SIGKDD CONFERENCE ON KNOWLEDGE DISCOVERY AND DATA MINING, P1027
  • [76] DTCDR: A Framework for Dual-Target Cross-Domain Recommendation
    Zhu, Feng
    Chen, Chaochao
    Wang, Yan
    Liu, Guanfeng
    Zheng, Xiaolin
    [J]. PROCEEDINGS OF THE 28TH ACM INTERNATIONAL CONFERENCE ON INFORMATION & KNOWLEDGE MANAGEMENT (CIKM '19), 2019, : 1533 - 1542
  • [77] Zhu QN, 2019, AAAI CONF ARTIF INTE, P5973
  • [78] Zhuang Fuzhen, 2010, Proceedings of the 19th ACM international conference on Information and knowledge management, P359