Practically Unbiased Pairwise Loss for Recommendation With Implicit Feedback

doi:10.1109/TPAMI.2024.3519711

	Practically Unbiased Pairwise Loss for Recommendation With Implicit Feedback
	Cao, Tianwei 1,2; Xu, Qianqian 3; Yang, Zhiyong 1; Ma, Zhanyu 2,4; Huang, Qingming 1,3,5,6
	2025-04-01
发表期刊	IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
ISSN	0162-8828
卷号	47 期号:4 页码:2460-2474
摘要	Recommender systems have been widely employed on various online platforms to improve user experience. In these systems, recommendation models are often learned from the users' historical behaviors that are automatically collected. Notably, recommender systems differ slightly from ordinary supervised learning tasks. In recommender systems, there is an exposure mechanism that decides which items could be presented to each specific user, which breaks the i.i.d assumption of supervised learning and brings biases into the recommendation models. In this paper, we focus on unbiased ranking loss weighted by inversed propensity scores (IPS), which are widely used in recommendations with implicit feedback labels. More specifically, we first highlight the fact that there is a gap between theory and practice in IPS-weighted unbiased loss. The existing pairwise loss could be theoretically unbiased by adopting an IPS weighting scheme. Unfortunately, the propensity scores are hard to estimate due to the inaccessibility of each user-item pair's true exposure status. In practical scenarios, we can only approximate the propensity scores. In this way, the theoretically unbiased loss would be still practically biased. To solve this problem, we first construct a theoretical framework to obtain a generalization upper bound of the current theoretically unbiased loss. The bound illustrates that we can ensure the theoretically unbiased loss's generalization ability if we lower its implementation loss and practical bias at the same time. To that aim, we suggest treating feedback label Y-ui as a noisy proxy for exposure result for O-ui each user-item pair (u,i). Here we assume the noise rate meets the condition that (P) over cap O-ui=1, Y-ui not equal O-ui) < 1/2. According to our analysis, this is a mild assumption that can be satisfied by many real-world applications. Based on this, we could train an accurate propensity model directly by leveraging a noise-resistant loss function. Then we could construct a practically unbiased recommendation model weighted by precise propensity scores. Lastly, experimental findings on public datasets demonstrate our suggested method's effectiveness.
关键词	Estimation Imputation Training data IP networks Accuracy Supervised learning Recommender systems Noise Data models Computer science Recommender system collaborative filtering debias noise-robust learning pairwise ranking loss
DOI	10.1109/TPAMI.2024.3519711
收录类别	SCI
语种	英语
资助项目	National Key R&D Program of China[2018AAA0102000] ; National Natural Science Foundation of China[62225601] ; National Natural Science Foundation of China[U23B2052] ; National Natural Science Foundation of China[62236008] ; National Natural Science Foundation of China[U21B2038] ; National Natural Science Foundation of China[U23B2051] ; National Natural Science Foundation of China[61931008] ; National Natural Science Foundation of China[62122075] ; National Natural Science Foundation of China[62206264] ; National Natural Science Foundation of China[92370102] ; Youth Innovative Research Team of BUPT[2023YQTD02] ; Youth Innovation Promotion Association CAS ; Strategic Priority Research Program of the Chinese Academy of Sciences[XDB0680000] ; Strategic Priority Research Program of the Chinese Academy of Sciences[XDB0680000] ; Innovation Funding of ICT, CAS[E000000] ; Beijing Natural Science Foundation[L242025] ; Hebei Natural Science Foundation[F2024502017]
WOS研究方向	Computer Science ; Engineering
WOS类目	Computer Science, Artificial Intelligence ; Engineering, Electrical & Electronic
WOS记录号	WOS:001439648900050
出版者	IEEE COMPUTER SOC
引用统计
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/40725
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Xu, Qianqian; Huang, Qingming
作者单位	1.Univ Chinese Acad Sci, Sch Comp Sci & Technol, Beijing 101408, Peoples R China 2.Beijing Univ Posts & Telecommun, Sch Artificial Intelligence, Pattern Recognit & Intelligent Syst Lab, Beijing 100876, Peoples R China 3.Chinese Acad Sci, Key Lab Intelligent Informat Proc, Inst Comp Technol, Beijing 100190, Peoples R China 4.Beijing Univ Posts & Telecommun, Sch Artificial Intelligence, Pattern Recognit & Intelligent Syst Lab, Beijing, Peoples R China 5.Univ Chinese Acad Sci, Key Lab Big Data Min & Knowledge Management BDKM, Beijing 101408, Peoples R China 6.Peng Cheng Lab, Shenzhen 518055, Peoples R China
推荐引用方式 GB/T 7714	Cao, Tianwei,Xu, Qianqian,Yang, Zhiyong,et al. Practically Unbiased Pairwise Loss for Recommendation With Implicit Feedback[J]. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE,2025,47(4):2460-2474.
APA	Cao, Tianwei,Xu, Qianqian,Yang, Zhiyong,Ma, Zhanyu,&Huang, Qingming.(2025).Practically Unbiased Pairwise Loss for Recommendation With Implicit Feedback.IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE,47(4),2460-2474.
MLA	Cao, Tianwei,et al."Practically Unbiased Pairwise Loss for Recommendation With Implicit Feedback".IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE 47.4(2025):2460-2474.