The DOI matches its Crossref record, with a small difference in title or year; 47 of the 48 references it cites were found in a registry or an index.
Popularity Bias is not Always Evil: Disentangling Benign and Harmful Bias for Recommendation
Open accessAbstract
Recommender system usually suffers from severepopularity bias— the collected interaction data usually exhibits quite imbalanced or even long-tailed distribution over items. Such skewed distribution may result from the users’conformityto the group, which deviates from reflecting users’ true preference. Existing efforts for tackling this issue mainly focus on completely eliminating popularity bias. However, we argue that not all popularity bias is evil. Popularity bias not only results from conformity but alsoitem quality, which is usually ignored by existing methods. Some items exhibit higher popularity as they have intrinsic better property. Blindly removing the popularity bias would lose such important signal, and further deteriorate model performance. To sufficiently exploit such important information for recommendation, it is essential to disentangle the benign popularity bias caused by item quality from the harmful popularity bias caused by conformity. Although important, it is quite challenging as we lack an explicit signal to differentiate the two factors of popularity bias. In this paper, we propose to leverage temporal information as the two factors exhibit quite different patterns along the time: item quality revealing item inherent property is stable and static while conformity that depends on items’ recent clicks is highly time-sensitive. Correspondingly, we further propose a novelTime-awareDisEntangled framework (TIDE), where a click is generated from three components namely the static item quality, the dynamic conformity effect, as well as the user-item matching score returned by any recommendation model. Lastly, we conduct interventional inference so that the recommendation can benefit from the benign popularity bias while circumvent the harmful one. Extensive experiments on four real-world datasets demonstrated the effectiveness of TIDE.
Overview
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“An appropriate proportion of popularity bias in the system can improve performance and user satisfaction [183].”
From A Deep Dive into Fairness, Bias, Threats, and Privacy in Recommender Systems: Insights and Future Research (2024) · cites this paper as [183]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Moreover, popularity bias can act as a double-edged sword [182, 183].”
From A Deep Dive into Fairness, Bias, Threats, and Privacy in Recommender Systems: Insights and Future Research (2024) · cites this paper as [183]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Auto-Encoding Variational Bayes
Kingma, D. P., & Welling, M. (2013) · arXiv · doi:10.48550/arxiv.1312.6114 · Cited by 20,067
Neural Collaborative Filtering
He, X., Liao, L., Zhang, H., Nie, L., Hu, X., & Chua, T. (2017) · International World Wide Web Conference (WWW) · doi:10.1145/3038912.3052569 · Cited by 7,385 · cited 1×
“experiments, and it would be straightforward to replace it with more sophisticated models such as Factorization Machine [15], or Neural Network [5], [16].”
MethodSelf-citationBPR: Bayesian Personalized Ranking from Implicit FeedbackPreprint
Rendle, S., Freudenthaler, C., Gantner, Z., & Schmidt-Thieme, L. (2012) · arXiv (Cornell University) · doi:10.48550/arxiv.1205.2618 · Cited by 6,168
He, X., Deng, K., Wang, X., Li, Y., Zhang, Y., & Wang, M. (2020) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3397271.3401063 · Cited by 5,300 · cited 4×
“Amazon-CDs is increased to 8,192 for speed and the number of convolutional layers is searched in {2, 3, 4} as advised by [5].”
BackgroundMethodInfluentialSelf-citationPearson Correlation Coefficient
Benesty, J., Chen, J., Huang, Y. A., & Cohen, I. (2009) · Springer topics in signal processing · doi:10.1007/978-3-642-00296-0_5 · Cited by 3,540
Self-citationThe long tail: why the future of business is selling less of more
(2007) · Choice Reviews Online · doi:10.5860/choice.44-2783 · Cited by 2,869
Deep Interest Network for Click-Through Rate Prediction
Zhou, G., Zhu, X., Song, C., Fan, Y., Zhu, H., Ma, X., Yan, Y., Jin, J., Li, H., & Gai, K. (2018) · Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining · doi:10.1145/3219819.3219823 · Cited by 2,128 · cited 2×
“(3) ( U, I ) → M → Y : these links project user and item features ( e.g., IDs) into their matching scores m ui = f θ ( u, i ) . f θ ( u, i ) can be implemented by various recommendation models, such as MF [4], LightGCN [5], DIN [6], etc. Finally these three components are aggregated into a final prediction score for r…”
BackgroundFactorization Machines with libFM
Rendle, S. (2012) · ACM Transactions on Intelligent Systems and Technology · doi:10.1145/2168752.2168771 · Cited by 1,345 · cited 1×
“Matrix Factorization (MF) has been selected as the main backbone recommendation model for experiments, and it would be straightforward to replace it with more sophisticated models such as Factorization Machine [15], or Neural Network [5], [16].”
MethodBias and Debias in Recommender System: A Survey and Future Directions
Chen, J., Dong, H., Wang, X., Feng, F., Wang, M., & He, X. (2023) · ACM Transactions on Information Systems · doi:10.1145/3564284 · Cited by 741
Self-citationDisentangled Graph Collaborative Filtering
Wang, X., Jin, H., Zhang, A., He, X., Xu, T., & Chua, T.-S. (2020) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3397271.3401137 · Cited by 737
Self-citation
Flagged first, then by citations
1 of the 48 references couldn’t be matched to a known paper.
Cited by
Sorted by citations
Items 101–117 of 117
A Multi-level Debiasing Recommendation Framework Based on Causal Disentanglement
Wang, S., Yang, J., Dong, Z., Xiong, J., Huang, W., & Yan, C. (2026) · Lecture notes in computer science · doi:10.1007/978-981-95-5640-3_3
Counterfactual reasoning for alleviating dual biases in recommendation
Lu, S., Luo, F., & Wu, J. (2026) · International Journal of Machine Learning and Cybernetics · doi:10.1007/s13042-026-03012-9
“This is because popularity bias and conformity bias are not entirely harmful and can be leveraged [ 6, 7 ].”
Training-Free and Unbiased Graph Collaborative Filtering for Personalized Recommendations
Liu, Z., Wang, C., Wu, C., et al. (2026) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2026.3669816
“Popularity bias [33], [34], [35] occurs when the varying popularity levels of items lead recommendation models to preferentially rank more popular items higher, irrespective of individual user preferences.”
TDCRec: Time-Varying Demand Causal Modeling for Recommendation Debiasing
Ma, J., Zhao, W., & Li, L. (2026) · Lecture notes in computer science · doi:10.1007/978-981-95-7075-1_31
Bias in Book Recommendation: A Case Study on the Danish Public Libraries
Daniil, S., Mollerup, S. H., & Laura Hollink, A. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-21324-2_22
Decomposing beneficial and harmful bias factors for conformity-aware causal recommendation
Tang, R., Yang, C., & Peng, Y. (2026) · Neurocomputing · doi:10.1016/j.neucom.2026.134022
FMCNS: Flow Matching with Causal-aware Negative Sampling for Multimodal Recommendation
Zha, X., Zha, Y., Jiang, H., et al. (2026) · Information Processing & Management · doi:10.1016/j.ipm.2026.104944
Beyond the latest: Correcting Release Interval Bias in short-video recommendation
Long, X., Zhang, H., Xu, Y.-Y., et al. (2026) · Knowledge-Based Systems · doi:10.1016/j.knosys.2026.116486
Hierarchical Dual Debiasing in LLM-based Recommendation
Lu, S., & Wu, J. (2026) · ACM Transactions on Information Systems · doi:10.1145/3837081
“(3) Causal inference-based methods [11, 44, 55, 61, 66] perform deconfounded training and conduct causal interventions during inference to alleviate the adverse impact of item popularity.”
Zhang, Z., & Zhang, Y. (2026) · Expert Systems with Applications · doi:10.1016/j.eswa.2026.133954
Li, Q., He, S., Zhao, Y., Chen, N., Wang, J., & Ma, Y. (2026) · ACM Transactions on the Web · doi:10.1145/3838705
DualPop: User Popularity Preference-Aware Debiasing Recommendation
Li, Z., Chi, Y., Shi, Y., Wang, J., Chen, Z., & Li, H. (2026) · ACM Transactions on Knowledge Discovery from Data · doi:10.1145/3833876
“In addition, we will further optimize the model structure to reduce computational overhead and develop a lightweight variant to balance recommendation performance and efficiency for practical deployment.”
Multi-behavior popularity-aware recommendation
Xu, Y.-Y., Chen, J.-Y., Zhang, W.-T., et al. (2026) · Information Fusion · doi:10.1016/j.inffus.2026.104771
Self-citationGTP: Mitigating Popularity Bias in PLM-based Sequential Recommendation via Group-Aware Token Pruning
Yuan, R., Liu, D., Chen, H., & Ming, Z. (2026) · ACM Conference on Recommender Systems (RecSys) · doi:10.1145/3773078.3831805
“Existing research mit-igates popularity bias primarily via Inverse Propensity Scoring (IPS) re-weighting [26, 28, 40] or causal disentanglement [6, 38].”
Cross-Spectrum Candidate Set Analysis for Popularity-Aware Recommendation
Jeong, H.-J., Lee, Y.-J., Kim, G.-W., et al. (2026) · ACM Conference on Recommender Systems (RecSys) · doi:10.1145/3773078.3841293
“Moreover, distinguishing beneficial popularity from that ∗ induced by conformity is difficult because the underlying cause is generally unobservable [10].”
Lapin, V. G., Gediev, K., Gordova, M., & Elashmawi, W. H. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-27834-0_28
Roy, F., Ding, X., Choo, K. K. R., & Zhou, P. (2024) · arXiv · doi:10.48550/arxiv.2409.12651 · Cited by 7
Metadata
- Authors
- Zhao, Z., Chen, J., Zhou, S., He, X., Cao, X., Zhang, F., & Wu, W.
- Affiliations
- University of Science and Technology of China (CN)
- Zhejiang University (CN)
- Meituan (China) (CN)
- Venue
- IEEE Transactions on Knowledge and Data Engineering
- Published
- 2023-10-01
- Volume
- 35
- Issue
- 10
- Pages
- 9920-9931
- ISSN
- 1041-4347 (ISSN-L)
- 1558-2191
- 2326-3865
- OpenAlex
- W4312551924
- Semantic Scholar
- 3d361ead1582071a5ed26cff391504746a1210a6
- Cited by
- 117
- References
- 48
- Citations by year
- 2026: 21 · 2025: 36 · 2024: 35 · 2023: 21
- FWCI
- 23.75
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- IEEE Transactions on Knowledge and Data Engineering (publishedVersion)
- arXiv (Cornell University) (submittedVersion, cc-by)
- arXiv (Cornell University) (cc-by)
- Funders
- National Key Research and Development Program of China (2021ZD0111802)
- National Natural Science Foundation of China (62102382, 61972372, U21B2026)
- Meituan Inc.
- CCCD Key Lab of Ministry of Culture and Tourism
- Starry Night Science Fund of Zhejiang University Shanghai Institute for Advanced Study (SN-ZJU-SIAS-001)
- Topics
- Physical Sciences › Computer Science › Information Systems › Recommender Systems and Techniques
- Physical Sciences › Computer Science › Information Systems › Information Retrieval and Search Behavior
- Social Sciences › Decision Sciences › Management Science and Operations Research › Advanced Bandit Algorithms Research
- Keywords
- popularity bias, recommender systems, item quality, temporal modeling, disentangled representations, click data
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
