Pith. sign in

REVIEW 4 major objections 6 minor 68 references

This paper claims that popularity bias in graph-neural-network recommenders can be removed after training by projecting node embeddings away from an estimated popularity direction, with no retraining.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-04 09:49 UTC pith:AIMZR5MK

load-bearing objection Post-hoc debiasing is a promising idea, but the popularity direction is a heuristic that needs independent validation before the reported gains can be trusted. the 4 major comments →

arxiv 2510.12959 v2 pith:AIMZR5MK submitted 2025-10-14 cs.IR

Post-hoc Popularity Bias Correction in GNN-based Collaborative Filtering

Md Aminul Islam , Elena Zheleva , Ren Wang This is my paper
classification cs.IR
keywords popularity biascollaborative filteringgraph neural networkspost-hoc debiasingembedding projectionrecommender systemsunbiased evaluationlong-tail distribution
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The authors are trying to establish that a post-hoc, training-free correction can undo popularity bias that graph neural networks amplify during message passing. They define an interaction-level popularity score for each user–item pair, combine it into per-node popularity and preference centroids, and use their difference as a popularity direction vector. Projecting each learned embedding onto that direction and subtracting the component removes popularity-related signal while keeping preference signals. If correct, deployed recommender systems could be debiased cheaply by editing embeddings after training, without retraining.

Core claim

PPD works directly on embeddings from a pre-trained GNN-based collaborative filtering model. For each interaction it estimates a popularity score b_ui = p_i − r_ui, where p_i is the item's average similarity to all users (global preference) and r_ui is the item's average similarity to the user's historical items, penalized by global preference. These scores yield two centroids per node, a popularity centroid and a preference centroid; the difference, scaled by coefficient phi, defines a per-node popularity direction vector. The node embedding is then updated by subtracting its vector projection onto that direction at layer 0, and the debiased embeddings are propagated through the remaining l

What carries the argument

The popularity direction vector d_pop(v) = e_bar_pop(v) − phi * e_bar_pref(v), built from interaction-level popularity scores b_ui = p_i − r_ui. Projection of each node embedding onto this direction and subtraction of that component is the mechanism claimed to strip popularity while preserving preference.

Load-bearing premise

The method assumes that the difference between the popularity centroid and the preference centroid points precisely along the popularity confound in embedding space, so that subtracting the projection removes popularity and nothing else.

What would settle it

Train a GNN-based CF model on a dataset with known unbiased ratings, construct random unit vectors in embedding space, and apply the same projection-and-subtraction step (tuning phi and beta on validation data); if random directions yield similar improvements in unbiased metrics, then the specific popularity-direction construction is not what carries the performance.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Deployed GNN recommender systems can be debiased without retraining, by editing the base embeddings once.
  • The method applies to any GNN-based CF backbone that produces user/item embeddings, not just one architecture.
  • On datasets with strong popularity skew, relative gains are large (e.g., the paper reports improvements over baselines on all metrics).
  • Head and tail item performance can improve together, rather than trading off.
  • Increasing GNN depth does not necessarily hurt when debiasing is applied, mitigating bias amplification and over-smoothing.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the projection direction is truly aligned with popularity, the same procedure could in principle be applied to embeddings from non-GNN recommender models, but the paper only tests GNN backbones.
  • The method's reliance on tunable hyperparameters phi and beta suggests the 'popularity direction' is not uniquely identifiable from the data alone; a validation set is used to pick them.
  • A testable extension: apply PPD to embeddings from matrix factorization or transformer-based recommenders to see whether the debiasing geometry transfers.
  • The paper leaves implicit that the projection could be composed with other post-hoc corrections (e.g., exposure or position bias) by defining analogous direction vectors.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes PPD, a post-hoc method for correcting popularity bias in GNN-based collaborative filtering. PPD first estimates an interaction-level popularity score b_ui = p_i - r_ui, where p_i is a global-preference measure and r_ui is a personalized-preference measure with a popularity penalty. It then constructs a per-node popularity direction d_pop(v) as the difference between a popularity centroid and a preference centroid, projects the layer-0 node embedding onto this direction, and subtracts the projected component before re-propagating through the GNN. Experiments on KuaiRec, Coat, and Yahoo! R3 with LightGCN and SGL backbones report consistent improvements over popularity-debiasing baselines, with especially large relative gains on KuaiRec.

Significance. If the proposed direction vector genuinely isolates popularity from preference, the method would be practically valuable because it debiases deployed embeddings without retraining and can be layered on any GNN-based CF backbone. The paper has strengths: it evaluates on three datasets with unbiased test sets, compares with a broad set of recent baselines, and includes ablations for the key hyperparameters and for a second backbone. However, the central construction is heuristic and is not validated as a popularity direction; the reported gains could in principle come from a tuned linear perturbation of the embeddings. The lack of error bars and significance tests makes it difficult to assess the small gains on Coat and Yahoo! R3. These issues are addressable with additional experiments and analysis.

major comments (4)
  1. [§4.1–4.2, Eqs. (7)–(12)] The central claim that the projection removes popularity rather than an arbitrary embedding direction is not supported. b_ui = p_i - r_ui can be negative, so the centroids in Eqs. (8)–(9) are not convex combinations and d_pop(v) is not guaranteed to be interpretable as a “popularity direction”; it is a data-dependent linear functional of the same biased embeddings it is later used to transform. The paper provides no diagnostic showing that d_pop(v) aligns with item popularity (e.g., degree or interaction frequency), and the success on the unbiased test could in principle come from a tuned linear perturbation. Please add: (i) correlation/alignment analysis between d_pop and popularity measures; (ii) a random-direction or fixed-direction control; (iii) error bars over seeds. This is necessary to substantiate the debiasing interpretation.
  2. [§5.2, Table 3 (Yahoo! R3, bottom 80% NDCG)] The text states “PPD consistently achieves the best performance across both groups.” In Table 3, for Yahoo! R3 bottom 80% NDCG@20, APDA reports 0.0129 while PPD reports 0.0127; PPD is not the best. The later sentence acknowledges a –1.5% drop, but the summary claim and the conclusion that PPD improves both head and tail recommendations should be qualified. This matters because RQ2 is specifically about the absence of a head–tail trade-off.
  3. [§5.1–5.2, Tables 2–4] No standard deviations, significance tests, or multiple-seed results are reported. The Coat and Yahoo gains over the best baseline are small (0.9–8%), while β and φ are tuned on a one-third split of the unbiased data. The reader cannot tell whether PPD is statistically better than the comparison methods or whether the gap is within noise. Report mean ± std over at least 5 seeds and paired significance tests (e.g., paired t-test or Wilcoxon) on the unbiased test folds.
  4. [§4.2, Eqs. (11)–(12)] Popularity scores in §4.1 are computed from the final/readout embeddings e_u, e_i (Section 3.1), but the projection is applied to the layer-0 embeddings e^(0). The relation between the final-embedding popularity direction and the layer-0 representation is not discussed. If the intention is to remove a direction estimated from the final embedding from the initial embedding, the calibration of this step needs justification and an experiment (e.g., applying the debiasing at every layer or at the final layer).
minor comments (6)
  1. [Algorithm 1, line 13] Typo: “preference centroid ¯e_pop(v)” should read “preference centroid ¯e_pref(v).”
  2. [Appendix B.2] The cross-reference to Figure 3 for RQ2 results is wrong; the Recall/NDCG results for RQ2 are in Table 3.
  3. [Table 2] IPSCN is cited as [21] in Table 2 but as [16] in Section 5.1 and Appendix B.1; the citation should be [16] (with [21] as the IPW source).
  4. [Eqs. (8)–(9)] The ε guard only prevents exact zero denominators; it does not address negative or near-zero sums when b_ui takes negative values. Please state the range of b_ui after min–max normalization and discuss the interpretation of negative weights.
  5. [Appendix A.2] The complexity simplification drops the |E| term because |E| << |U||I|, but the retained term is |E| * average degree * d; the justification is incomplete. Also, the proposed user sampling to reduce O(|U||I|d) is not used in the experiments.
  6. [§5.2 RQ2 text] The phrase “PPD consistently achieves the best performance across both groups” is too strong given the Yahoo tail NDCG result (see Major 2). Please rephrase.

Circularity Check

0 steps flagged

No significant circularity: the popularity direction is an unvalidated heuristic rather than a prediction forced by construction, and the paper's central empirical claims are checked on held-out unbiased test data.

full rationale

The derivation chain does not reduce to its inputs in the sense prohibited by the rubric. The popularity score b_ui = p_i - r_ui (Eq. 7) is an operational definition of popularity, not a parameter fitted to the unbiased test labels and then renamed as a prediction. The popularity direction d_pop = e_pop - phi*e_pref (Eq. 10) is built from the same pre-trained embeddings that are later projected (Eqs. 11-12), making PPD a data-dependent linear transformation; whether this direction truly aligns with the popularity confound is an assumption, not a derived theorem. The paper itself acknowledges this by saying the adjustment can only be 'indirectly evaluated through improvements in predictive accuracy through an unbiased evaluation' (Sec. 3.2). An unvalidated assumption is a correctness risk, not a circular reduction: the unbiased test splits (Sec. 5.1) are held out and serve as an external benchmark independent of the method's popularity-score definition. Hyperparameters beta and phi are tuned on a one-third validation split of the unbiased data and evaluated on the remaining two-thirds, which is standard model selection rather than forced prediction. The only self-citation is ref. [33] in the related-work discussion of IPW propensity estimation; it is not load-bearing for the proposed method. No uniqueness theorems, no ansatz smuggled in via self-citation, and no renaming of a known result are present. The claimed improvements are empirical, and the debiasing mechanism is a heuristic whose semantic interpretation should be validated further, but this is not circularity.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 2 invented entities

The central claim rests on the assumption that the embedding-space geometry is interpretable enough that a linear direction can separate popularity from preference. The only free parameters are two per-dataset hyperparameters plus a division guard. The 'popularity direction' is an invented construct with no independent evidence tying it to actual popularity.

free parameters (3)
  • β (popularity penalty coefficient) = KuaiRec ≈ 0.2; Coat ≈ 0.1; Yahoo! R3 ≈ 0.3 (from Fig. 4)
    Controls the penalty on global preference inside personalized preference (Eq. 6); tuned per dataset on the unbiased validation split.
  • φ (preference centroid coefficient) = KuaiRec ≈ 0.50; Coat ≈ 1.0; Yahoo! R3 ≈ 1.0 (from Fig. 3)
    Scales the preference centroid when defining the popularity direction (Eq. 10); tuned per dataset on the unbiased validation split.
  • ε (division guard) = 1e-8
    Small constant to avoid division by zero in Eqs. (8)–(9); not a meaningful free parameter.
axioms (5)
  • domain assumption Embedding geometry encodes popularity and preference as separable directions; the vector from preference centroid to popularity centroid captures the popularity confound.
    The method's validity depends on this without independent verification; no ground-truth popularity direction is available.
  • domain assumption Popularity and preference signals are linearly separable, so removing only the one-dimensional projection along d_pop leaves preference information intact.
    Assumed in Eqs. (11)–(12); the paper does not test whether projection removes preference-related signal.
  • domain assumption The unbiased test sets (KuaiRec dense subset, Coat/Yahoo random ratings) provide a reliable counterfactual evaluation.
    All empirical conclusions depend on the MAR assumption of these test sets, which is standard in the field but not verified here.
  • domain assumption Pre-trained BPR-trained embeddings are stable enough for post-hoc popularity estimation.
    The method uses embeddings from a converged model; if embeddings are noisy or oversmoothed, the estimated popularity scores may be unreliable.
  • standard math Standard vector projection formula is valid linear algebra.
    Eqs. (11)–(12) use the standard orthogonal projection of one vector onto another.
invented entities (2)
  • Popularity direction vector d_pop(v) no independent evidence
    purpose: Defines the direction in embedding space along which popularity is assumed to pull node representations; PPD removes the projection of each embedding onto this vector.
    The vector is constructed from weighted centroids of the same embeddings being debiased; no external measure of popularity validates that this direction encodes popularity rather than arbitrary geometry.
  • Interaction-level popularity score b_ui no independent evidence
    purpose: Quantifies how much each user–item interaction is driven by item popularity instead of user preference; used to weight the centroids.
    Defined as p_i − r_ui, both quantities computed from the embedding space; no direct validation against known popularity labels or causal effects.

pith-pipeline@v1.3.0-alltime-deepseek · 20603 in / 10554 out tokens · 89959 ms · 2026-08-04T09:49:04.040512+00:00 · methodology

0 comments
read the original abstract

User historical interaction data is the primary signal for learning user preferences in collaborative filtering (CF). However, the training data often exhibits a long-tailed distribution, where only a few items have the majority of interactions. CF models trained directly on such imbalanced data are prone to learning popularity bias, which reduces personalization and leads to suboptimal recommendation quality. Graph Neural Networks (GNNs), while effective for CF due to their message passing mechanism, can further propagate and amplify popularity bias through their aggregation process. Existing approaches typically address popularity bias by modifying training objectives but fail to directly counteract the bias propagated during GNN's neighborhood aggregation. Applying weights to interactions during aggregation can help alleviate this problem, yet it risks distorting model learning due to unstable node representations in the early stages of training. In this paper, we propose a Post-hoc Popularity Debiasing (PPD) method that corrects for popularity bias in GNN-based CF and operates directly on pre-trained embeddings without requiring retraining. By estimating interaction-level popularity and removing popularity components from node representations via a popularity direction vector, PPD reduces bias while preserving user preferences. Experimental results show that our method outperforms state-of-the-art approaches for popularity bias correction in GNN-based CF.

Figures

Figures reproduced from arXiv: 2510.12959 by Elena Zheleva, Md Aminul Islam, Ren Wang.

Figure 1
Figure 1. Figure 1: Distribution of interactions across head (top [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Effect of varying the number of layers on performance (Recall@ [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Effect of varying the preference centroid coefficient ( [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Effect of varying the popularity penalty coefficient ( [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Effect of varying the number of layers on performance (NDCG@ [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

68 extracted references · 1 linked inside Pith

  1. [1]

    Himan Abdollahpouri, Robin Burke, and Bamshad Mobasher. 2019. Managing popularity bias in recommender systems with personalized re-ranking. InFLAIRS

    2019

  2. [2]

    Qingyao Ai, Keping Bi, Cheng Luo, Jiafeng Guo, and W Bruce Croft. 2018. Unbi- ased learning to rank with unbiased propensity estimation. InThe 41st interna- tional ACM SIGIR conference on research & development in information retrieval. 385–394

    2018

  3. [3]

    Stephen Bonner and Flavian Vasile. 2018. Causal embeddings for recommendation. InProceedings of the 12th ACM conference on recommender systems. 104–112

    2018

  4. [4]

    Ludovico Boratto, Gianni Fenu, and Mirko Marras. 2021. Connecting user and item perspectives in popularity debiasing for collaborative recommendation. Information Processing & Management58, 1 (2021), 102387

    2021

  5. [5]

    Léon Bottou, Jonas Peters, Joaquin Quiñonero-Candela, Denis X Charles, D Max Chickering, Elon Portugaly, Dipankar Ray, Patrice Simard, and Ed Snelson. 2013. Counterfactual reasoning and learning systems: The example of computational advertising.The Journal of Machine Learning Research14, 1 (2013), 3207–3260

    2013

  6. [6]

    Rocío Cañamares and Pablo Castells. 2018. Should I follow the crowd? A prob- abilistic analysis of the effectiveness of popularity in recommender systems. InThe 41st International ACM SIGIR Conference on Research & Development in Information Retrieval. 415–424

    2018

  7. [7]

    Allison JB Chaney, Brandon M Stewart, and Barbara E Engelhardt. 2018. How algorithmic confounding in recommendation systems increases homogeneity and decreases utility. InProceedings of the 12th ACM conference on recommender systems. 224–232

    2018

  8. [8]

    Deli Chen, Yankai Lin, Wei Li, Peng Li, Jie Zhou, and Xu Sun. 2020. Measuring and relieving the over-smoothing problem for graph neural networks from the topological view. InProceedings of the AAAI conference on artificial intelligence, Vol. 34. 3438–3445

    2020

  9. [9]

    Hao Chen, Zefan Wang, Feiran Huang, Xiao Huang, Yue Xu, Yishi Lin, Peng He, and Zhoujun Li. 2022. Generative adversarial framework for cold-start item recommendation. InProceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2565–2571

    2022

  10. [10]

    Hao Chen, Yue Xu, Feiran Huang, Zengde Deng, Wenbing Huang, Senzhang Wang, Peng He, and Zhoujun Li. 2020. Label-aware graph convolutional net- works. InProceedings of the 29th ACM international conference on information & knowledge management. 1977–1980

    2020

  11. [11]

    Jiajia Chen, Jiancan Wu, Jiawei Chen, Xin Xin, Yong Li, and Xiangnan He. 2024. How graph convolutions amplify popularity bias for recommendation?Frontiers of Computer Science18, 5 (2024), 185603

    2024

  12. [12]

    Zhihong Chen, Rong Xiao, Chenliang Li, Gangfeng Ye, Haochuan Sun, and Hongbo Deng. 2020. Esam: Discriminative domain adaptation with non-displayed items to improve long-tail performance. InProceedings of the 43rd international ACM SIGIR conference on research and development in information retrieval. 579– 588

    2020

  13. [13]

    Junnan Dong, Qinggang Zhang, Xiao Huang, Keyu Duan, Qiaoyu Tan, and Zhimeng Jiang. 2023. Hierarchy-aware multi-hop question answering over knowledge graphs. InProceedings of the ACM web conference 2023. 2519–2527

    2023

  14. [14]

    Chongming Gao, Shijun Li, Wenqiang Lei, Jiawei Chen, Biao Li, Peng Jiang, Xiangnan He, Jiaxin Mao, and Tat-Seng Chua. 2022. KuaiRec: A fully-observed dataset and insights for evaluating recommender systems. InProceedings of the 31st ACM International Conference on Information & Knowledge Management. 540–550

    2022

  15. [15]

    Chen Gao, Xiang Wang, Xiangnan He, and Yong Li. 2022. Graph neural net- works for recommender system. InProceedings of the fifteenth ACM international conference on web search and data mining. 1623–1625

    2022

  16. [16]

    Alois Gruson, Praveen Chandar, Christophe Charbuillet, James McInerney, Samantha Hansen, Damien Tardieu, and Ben Carterette. 2019. Offline evaluation to make decisions about playlist recommendation algorithms. InProceedings of the Twelfth ACM International Conference on Web Search and Data Mining. 420–428

    2019

  17. [17]

    Ming He, Changshu Li, Xinlei Hu, Xin Chen, and Jiwen Wang. 2022. Mitigating popularity bias in recommendation via counterfactual inference. InInternational Conference on Database Systems for Advanced Applications. Springer, 377–388

    2022

  18. [18]

    Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. Lightgcn: Simplifying and powering graph convolution network for recommendation. InProceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval. 639–648

    2020

  19. [19]

    Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural collaborative filtering. InProceedings of the 26th international conference on world wide web. 173–182

    2017

  20. [20]

    Zhongyu Huang, Yingheng Wang, Chaozhuo Li, and Huiguang He. 2022. Go- ing deeper into permutation-sensitive graph neural networks. InInternational conference on machine learning. PMLR, 9377–9409

    2022

  21. [21]

    Thorsten Joachims, Adith Swaminathan, and Tobias Schnabel. 2017. Unbiased learning-to-rank with biased feedback. InProceedings of the tenth ACM interna- tional conference on web search and data mining. 781–789

    2017

  22. [22]

    Toshihiro Kamishima, Shotaro Akaho, Hideki Asoh, and Jun Sakuma. 2014. Cor- recting popularity bias by enhancing recommendation neutrality.RecSys posters 10 (2014)

    2014

  23. [23]

    Minseok Kim, Jinoh Oh, Jaeyoung Do, and Sungjin Lee. 2022. Debiasing neighbor aggregation for graph neural network in recommender systems. InProceedings of the 31st ACM International Conference on Information & Knowledge Management. 4128–4132

    2022

  24. [24]

    Kipf and Max Welling

    Thomas N. Kipf and Max Welling. 2017. Semi-Supervised Classification with Graph Convolutional Networks. InThe Fifth International Conference on Learning Representations

    2017

  25. [25]

    Walid Krichene and Steffen Rendle. 2020. On sampled metrics for item recom- mendation. InProceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining. 1748–1757

    2020

  26. [26]

    Zihan Lin, Changxin Tian, Yupeng Hou, and Wayne Xin Zhao. 2022. Improving graph collaborative filtering with neighborhood-enriched contrastive learning. InProceedings of the ACM web conference 2022. 2320–2329

    2022

  27. [27]

    Dan Luo, Lixin Zou, Qingyao Ai, Zhiyu Chen, Chenliang Li, Dawei Yin, and Brian D Davison. 2024. Unbiased Learning-to-Rank Needs Unconfounded Propen- sity Estimation. InProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval. 1535–1545

    2024

  28. [28]

    Dan Luo, Lixin Zou, Qingyao Ai, Zhiyu Chen, Dawei Yin, and Brian D Davison

  29. [29]

    Kelong Mao, Jieming Zhu, Xi Xiao, Biao Lu, Zhaowei Wang, and Xiuqiang He

  30. [30]

    Benjamin M Marlin and Richard S Zemel. 2009. Collaborative prediction and ranking with non-random missing data. InProceedings of the third ACM conference on Recommender systems. 5–12

    2009

  31. [31]

    Wentao Ning, Reynold Cheng, Xiao Yan, Ben Kao, Nan Huo, Nur Al Hasan Haldar, and Bo Tang. 2024. Debiasing recommendation with personal popularity. In Proceedings of the ACM Web Conference 2024. 3400–3409

    2024

  32. [32]

    Harrie Oosterhuis and Maarten de Rijke. 2020. Policy-Aware Unbiased Learning to Rank for Top-k Rankings. InProceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. Association for Computing Machinery, 489–498

    2020

  33. [33]

    Zohreh Ovaisi, Kathryn Vasilaky, and Elena Zheleva. 2021. Propensity- independent bias recovery in offline learning-to-rank systems. InProceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. 1763–1767

    2021

  34. [34]

    2009.Geometric algebra with applications in engineering

    Christian Perwass. 2009.Geometric algebra with applications in engineering. Springer

    2009

  35. [35]

    Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme

  36. [36]

    Wondo Rhee, Sung Min Cho, and Bongwon Suh. 2022. Countering popularity bias by regularizing score differences. InProceedings of the 16th ACM conference on recommender systems. 145–155

    2022

  37. [37]

    Tobias Schnabel, Adith Swaminathan, Ashudeep Singh, Navin Chandak, and Thorsten Joachims. 2016. Recommendations as treatments: Debiasing learning and evaluation. Ininternational conference on machine learning. PMLR, 1670– 1679

    2016

  38. [38]

    Harald Steck. 2018. Calibrated recommendations. InProceedings of the 12th ACM conference on recommender systems. 154–162

    2018

  39. [39]

    Xiaoyuan Su and Taghi M Khoshgoftaar. 2009. A survey of collaborative filtering techniques.Advances in artificial intelligence2009, 1 (2009), 421425

    2009

  40. [40]

    Jianing Sun, Yingxue Zhang, Wei Guo, Huifeng Guo, Ruiming Tang, Xiuqiang He, Chen Ma, and Mark Coates. 2020. Neighbor interaction aware graph convolution networks for recommendation. InProceedings of the 43rd international ACM SIGIR conference on research and development in information retrieval. 1289–1298

    2020

  41. [41]

    Chenyang Wang, Yuanqing Yu, Weizhi Ma, Min Zhang, Chong Chen, Yiqun Liu, and Shaoping Ma. 2022. Towards representation alignment and uniformity in collaborative filtering. InProceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining. 1816–1825

    2022

  42. [42]

    Wenjie Wang, Fuli Feng, Xiangnan He, Xiang Wang, and Tat-Seng Chua. 2021. Deconfounded recommendation for alleviating bias amplification. InProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. 1717– 1725

    2021

  43. [43]

    Xuanhui Wang, Michael Bendersky, Donald Metzler, and Marc Najork. 2016. Learning to Rank with Selection Bias in Personal Search. InProceedings of the 39th International ACM SIGIR Conference on Research and Development in Information Retrieval. Association for Computing Machinery, New York, NY, USA, 115–124

    2016

  44. [44]

    Xiang Wang, Xiangnan He, Meng Wang, Fuli Feng, and Tat-Seng Chua. 2019. Neural graph collaborative filtering. InProceedings of the 42nd international ACM SIGIR conference on Research and development in Information Retrieval. 165–174

    2019

  45. [45]

    Xiang Wang, Hongye Jin, An Zhang, Xiangnan He, Tong Xu, and Tat-Seng Chua. 2020. Disentangled graph collaborative filtering. InProceedings of the 43rd 10 international ACM SIGIR conference on research and development in information retrieval. 1001–1010

    2020

  46. [46]

    Jacek Wasilewski and Neil Hurley. 2016. Incorporating Diversity in a Learning to Rank Recommender System.. InFLAIRS. 572–578

    2016

  47. [47]

    Tianxin Wei, Fuli Feng, Jiawei Chen, Ziwei Wu, Jinfeng Yi, and Xiangnan He

  48. [48]

    Jiancan Wu, Xiang Wang, Fuli Feng, Xiangnan He, Liang Chen, Jianxun Lian, and Xing Xie. 2021. Self-supervised graph learning for recommendation. InProceed- ings of the 44th international ACM SIGIR conference on research and development in information retrieval. 726–735

    2021

  49. [49]

    Kun Wu, Jie Shen, Yue Ning, Ting Wang, and Wendy Hui Wang. 2023. Certified edge unlearning for graph neural networks. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 2606–2617

    2023

  50. [50]

    Yonghui Yang, Le Wu, Richang Hong, Kun Zhang, and Meng Wang. 2021. En- hanced graph learning for collaborative filtering via mutual information maxi- mization. InProceedings of the 44th international ACM SIGIR conference on research and development in information retrieval. 71–80

    2021

  51. [51]

    InProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining

    Model-agnostic counterfactual reasoning for eliminating popularity bias in recommender system. InProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. 1791–1800

  52. [52]

    Junliang Yu, Xin Xia, Tong Chen, Lizhen Cui, Nguyen Quoc Viet Hung, and Hongzhi Yin. 2023. XSimGCL: Towards extremely simple graph contrastive learning for recommendation.IEEE Transactions on Knowledge and Data Engi- neering36, 2 (2023), 913–926

    2023

  53. [53]

    Junliang Yu, Hongzhi Yin, Xin Xia, Tong Chen, Lizhen Cui, and Quoc Viet Hung Nguyen. 2022. Are graph augmentations necessary? simple graph contrastive learning for recommendation. InProceedings of the 45th international ACM SIGIR conference on research and development in information retrieval. 1294–1303

    2022

  54. [54]

    Daochen Zha, Louis Feng, Bhargav Bhushanam, Dhruv Choudhary, Jade Nie, Yuandong Tian, Jay Chae, Yinbin Ma, Arun Kejariwal, and Xia Hu. 2022. Au- toshard: Automated embedding table sharding for recommender systems. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 4461–4471

    2022

  55. [55]

    Sirui Yao and Bert Huang. 2017. Beyond parity: Fairness objectives for collabora- tive filtering.Advances in neural information processing systems30 (2017)

    2017

  56. [56]

    An Zhang, Leheng Sheng, Zhibo Cai, Xiang Wang, and Tat-Seng Chua. 2023. Empowering collaborative filtering with principled adversarial contrastive loss. Advances in Neural Information Processing Systems36 (2023), 6242–6266

    2023

  57. [57]

    Peiyan Zhang, Jiayan Guo, Chaozhuo Li, Yueqi Xie, Jae Boum Kim, Yan Zhang, Xing Xie, Haohan Wang, and Sunghun Kim. 2023. Efficiently leveraging multi- level user intent for session-based recommendation via atten-mixer network. In Proceedings of the sixteenth ACM international conference on web search and data mining. 168–176

    2023

  58. [58]

    Yang Zhang, Fuli Feng, Xiangnan He, Tianxin Wei, Chonggang Song, Guohui Ling, and Yongdong Zhang. 2021. Causal intervention for leveraging popularity bias in recommendation. InProceedings of the 44th international ACM SIGIR conference on research and development in information retrieval. 11–20

    2021

  59. [59]

    An Zhang, Wenchang Ma, Xiang Wang, and Tat-Seng Chua. 2022. Incorporating bias-aware margins into contrastive loss for collaborative filtering.Advances in Neural Information Processing Systems35 (2022), 7866–7878

    2022

  60. [60]

    Chu Zhao, Enneng Yang, Yuliang Liang, Pengxiang Lan, Yuting Liu, Jianzhe Zhao, Guibing Guo, and Xingwei Wang. 2025. Graph representation learning via causal diffusion for out-of-distribution recommendation. InProceedings of the ACM on Web Conference 2025. 334–346

    2025

  61. [61]

    Chu Zhao, Enneng Yang, Yuliang Liang, Jianzhe Zhao, Guibing Guo, and Xingwei Wang. 2025. Distributionally robust graph out-of-distribution recommendation via diffusion model. InProceedings of the ACM on Web Conference 2025. 2018–2031

    2025

  62. [62]

    Zihao Zhao, Jiawei Chen, Sheng Zhou, Xiangnan He, Xuezhi Cao, Fuzheng Zhang, and Wei Wu. 2022. Popularity bias is not always evil: Disentangling benign and harmful bias for recommendation.IEEE Transactions on Knowledge and Data Engineering35, 10 (2022), 9920–9931

    2022

  63. [63]

    Yiding Zhang, Chaozhuo Li, Xing Xie, Xiao Wang, Chuan Shi, Yuming Liu, Hao Sun, Liangjie Zhang, Weiwei Deng, and Qi Zhang. 2022. Geometric disentan- gled collaborative filtering. InProceedings of the 45th international ACM SIGIR conference on research and development in information retrieval. 80–90

    2022

  64. [64]

    Ziwei Zhu, Yun He, Xing Zhao, and James Caverlee. 2021. Popularity bias in dynamic recommendation. InProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. 2439–2449. A More Details of Our Method A.1 PPD Algorithm The complete procedure of our proposed PPD method is outlined in Algorithm 1. The algorithm first estimates intera...

    2021

  65. [67]

    Huachi Zhou, Hao Chen, Junnan Dong, Daochen Zha, Chuang Zhou, and Xiao Huang. 2023. Adaptive popularity debiasing aggregator for graph collaborative filtering. InProceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval. 7–17

    2023

  66. [2012]

    BPR: Bayesian personalized ranking from implicit feedback.arXiv preprint arXiv:1205.2618(2012)

    Pith/arXiv arXiv 2012

  67. [2021]

    InProceedings of the 30th ACM international conference on information & knowledge management

    UltraGCN: ultra simplification of graph convolutional networks for recom- mendation. InProceedings of the 30th ACM international conference on information & knowledge management. 1253–1262

  68. [2023]

    InProceedings of the Sixteenth ACM International Conference on Web Search and Data Mining

    Model-based unbiased learning to rank. InProceedings of the Sixteenth ACM International Conference on Web Search and Data Mining. 895–903