Pith. sign in

REVIEW 5 major objections 5 minor 53 references

Improving Recommendation Fairness via Graph Structure and Representation Augmentation

T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read FairDDA improves recommendation fairness by pruning sensitive edges and masking sensitive features without utility loss.

desk verdict A sensible dual augmentation framework for fair graph recommenders, but a sign/terminology bug in the edge-pruning formula inverts the stated mechanism; worth a look because the code is available and the fix is likely a typo. read the letter →

arxiv 2508.19547 v1 pith:5HJGP5X6 submitted 2025-08-27 cs.IR

Tongxin Xu , Wenqiang Liu , Chenzhong Bin , Cihan Xiao , Zhixin Zeng , Tianlong Gu This is my paper · ORCID
classification cs.IR
keywords recommendersystemsgroupfairnessdataaugmentationgraphneuralnetworkssensitiveattributedemographicparityequalopportunityHilbert-Schmidtindependencecriterion
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes FairDDA, a method for making graph-based recommender systems fairer without sacrificing recommendation quality. Its central goal is to identify and remove sensitive information at the data level, by pruning user–item edges that carry bias and masking feature dimensions correlated with sensitive attributes, before training the final model. FairDDA is tested on two real-world datasets and is reported to lower Demographic Parity and Equal Opportunity scores while keeping or slightly improving recall and NDCG relative to the base graph recommender and several fairness baselines. This is a practical claim that fairness can be treated as a data-quality problem, addressed by targeted augmentation, rather than only as an added objective during representation learning.

What carries the argument

The load-bearing mechanism is a pair of augmentation operations driven by two prior hypotheses, together with a debiasing objective. Sensitive edge pruning constructs a masking matrix B by sampling each edge with probability $p_{uv} = \exp(\Delta r^d_u(v) - \Delta r^p_u(v))$, where $\Delta r^d$ and $\Delta r^p$ are the relative rankings of item $v$ in the debiased versus performance-oriented recommendation lists; the sampling is made differentiable by a Gumbel-softmax approximation. Sensitive feature masking constructs masks $f = \exp(-\sigma(d_\theta(x_d \odot x_b)))$ from the elementwise interaction of debiased and biased representations. The final representation is trained with BPR loss, a reconstruction loss, a contrastive loss that aligns original and augmented views, and an HSIC term that drives augmented user representations to be independent of biased ones.

What would settle it

Log the edges that get pruned during training and check the sign of $\Delta r^d_u(v) - \Delta r^p_u(v)$: Hypothesis 1 predicts that pruned edges are those whose item rank falls in the debiased ranking, while Eq. (10) as written predicts that pruned edges are those whose rank rises. If a run prunes mostly rising items, the mechanism is not acting as described. A second check is to flip the sign in Eq. (10) and rerun the ML-1M experiment: if the fairness improvement disappears or utility drops, the reported result depends on the sign of this score difference.

Watch

Extended reading notes

Core claim

FairDDA's central claim is that a dual data-augmentation schedule, sensitive edge pruning plus sensitive feature masking, can reduce group-level recommendation bias while preserving user preference information. The method first trains a performance-oriented model and a biased model that explicitly predicts users' sensitive attributes, then trains a debiased model. Guided by two hypotheses, it computes for each interaction the probability that the edge is sensitive from the difference between the debiased and performance-oriented rankings, and for each feature dimension the probability that the dimension is sensitive from the similarity of biased and debiased representations. The augmented graph and augmented features are then used with a contrastive consistency loss and an HSIC-based independence objective. The paper reports that this combination outperforms the base model and eight fairness baselines on ML-1M and LastFM, improving fairness substantially and retaining or slightly improving utility.

Load-bearing premise

The method relies on the formula $p_{uv} = \exp(\Delta r^d_u(v) - \Delta r^p_u(v))$ assigning the highest pruning probability to genuinely sensitive edges; if the sign convention is wrong, the augmentation removes the wrong interactions and the fairness gain cannot be attributed to the stated mechanism.

Editorial extensions

If this is right

  • Fairness improvement is compatible with recommendation accuracy: on the reported datasets, FairDDA keeps NDCG and Recall at or above the base LightGCN model while lowering DP and EO.
  • Data-level augmentation and representation-level debiasing are complementary; the ablation shows that removing either the edge-pruning strategy or the feature-masking strategy degrades fairness.
  • The approach transfers to multi-class sensitive attributes; on ML-1M with occupation, FairDDA again lowers DP and EO while keeping utility.
  • The best trade-off has a usable operating point: on ML-1M, $\lambda_r = 1$, $\lambda_c = 0.1$, and $\lambda_d = 30$ give the reported balance, with $\lambda_d$ controlling fairness strength.
  • Contrastive and reconstruction losses are what protect utility; without them the augmented graph would become an information-destroying filter.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the mechanism is as described, the same augmentation recipe could be applied to other graph-based prediction tasks where sensitive attributes flow through edges, such as link prediction in social networks, after replacing the recommendation loss with the task's own objective.
  • Because the pruning probability is a closed form based on rank differences, a parameter-free variant that thresholds the rank difference rather than exponentiating it would test whether the exact functional form matters or only the ordering of sensitive edges.
  • The current study measures user-side group fairness for binary gender and a multi-class occupation attribute; testing on intersectional groups, such as gender combined with occupation, would reveal whether the augmentation removes bias that accumulates at intersections or only the dominant group-level correlation.
  • The reliance on a fixed performance-oriented pretrained model means the quality of the fairness signal is bounded by that model's ranking quality; retraining the same pipeline on different backbones would show how sensitive the method is to the pretraining stage.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes FairDDA, a fairness-aware dual data augmentation framework for graph-based recommendation. FairDDA builds three user/item representation views (performance-oriented, biased, and debiased), then augments the interaction graph by pruning sensitive edges and masks sensitive feature dimensions in the input representations. Two prior hypotheses are proposed to identify sensitive edges (via ranking differences between performance-oriented and debiased recommendations) and sensitive features (via similarity between biased and debiased representations). The debiased representation is trained with a BPR objective, an HSIC-based debiasing loss, and contrastive/reconstruction losses. Experiments on ML-1M and LastFM report improved Demographic Parity and Equal Opportunity with utility comparable to or better than LightGCN and several fairness baselines, along with ablations, hyperparameter studies, and a multi-class sensitive attribute experiment.

Significance. If the method works as described, the paper addresses an important limitation of prior fairness-aware recommendation methods, namely that they often operate only at the objective level and neglect biased data distributions. The data-augmentation perspective, with explicit definitions of sensitive edges and features, is a reasonable and potentially useful direction. The paper also ships source code, reports ablations, and evaluates on multiple datasets and a multi-class sensitive attribute setting, which are strengths. However, the central mechanism as written contains sign inconsistencies that invert the stated hypotheses, and the statistical significance claims are contradicted by the reported p-values. These issues must be resolved before the contributions can be assessed reliably.

major comments (5)
  1. [Sec. 4.4.1, Eq. (10)-(11), vs. Hypothesis 1 (Sec. 4.2)] The pruning probability is defined as p_uv = exp(Δr^d_u(v) − Δr^p_u(v)) and explicitly called "the probability that the edge (u,v) is pruned from G". Hypothesis 1 states that an edge is sensitive when the item is ranked lower by the debiased model than by the performance-oriented model, i.e., Δr^d_u(v) − Δr^p_u(v) < 0. For such sensitive edges, Eq. (10) yields p_uv < 1, i.e., a low pruning probability, while for non-sensitive edges (difference > 0) p_uv is clipped to 1, i.e., pruned with certainty. This is exactly the inverse of the stated mechanism. Moreover, A^a = A ⊙ B with B_uv = 1 retaining the edge means a high p_uv corresponds to retention, not pruning. The equations are coherent only if p_uv is a retention probability, which contradicts the text. Since the ablation in Table 3 shows that the edge-pruning component contributes to fairness, this ambiguity is load-bearing and must be corrected.
  2. [Sec. 4.4.2, Eqs. (13)-(14)] The "sensitive feature masking" strategy does not mask as written. Eq. (13) sets X_a = X_d + X_d ⊙ F, and Eq. (14) defines F entries as exp(−σ(d_θ(x_d ⊙ x_b))). Since σ(·) ∈ (0,1), every entry of F lies in (e^{−1}, 1), so every feature is multiplied by a factor at least 1.367. The augmented representation is thus amplified relative to X_d, not masked. If the intended operation is to suppress sensitive features, the combination should be something like X_a = X_d ⊙ (1 − F) (or X_a = X_d − X_d ⊙ F), and the direction of Hypothesis 2 (high similarity should correspond to stronger suppression) must be aligned with the output of the detector. As written, the representation augmentation cannot be attributed to the stated sensitive-feature masking mechanism.
  3. [Sec. 5.2, Table 2] The text claims that "all improvements are statistically significant (p < 0.05)", but the reported p-values directly contradict this. In Table 2, the p-values for ML-1M Recall@20 (6.1e−2), LastFM NDCG@20 (3.4e−1), LastFM Recall@10 (6.8e−1), and LastFM Recall@30 (4.5e−1) are all above 0.05. In addition, FairDDA's LastFM NDCG@20 (0.2470) is lower than FairCoRe (0.2474), and its LastFM Recall@10 (0.1569) is lower than both FairMI (0.1576) and FairCoRe (0.1574). The paper should report which specific comparisons are statistically significant and should qualify the utility-preservation claim accordingly.
  4. [Sec. 4.4.1, Eq. (12)] The Gumbel-softmax approximation for the Bernoulli sampling is not correct as written. The standard continuous relaxation of a Bernoulli(p) variable uses the logit log(p/(1−p)), not log(p). With p clipped to 1 (Eq. (11)), log(p) = 0, so the Gumbel noise still leads to nondeterministic outputs, which does not respect the intended probability; with p = 0, log(p) is undefined. This means the actual edge-pruning distribution implemented in the model differs from the stated Bernoulli(p_uv), and it is unclear how the forward/backward approximation behaves. The authors should either use the proper binary Gumbel-softmax with logits or justify the alternative relaxation.
  5. [Sec. 4.5, Eq. (15) and Sec. 4.7.2] The HSIC definition uses m as "the number of user-item interactions retained after the sensitive edge pruning", but the kernel matrices K_a and K_b are computed over user representations X_a_U and X_b_U, which are indexed by users, not by interactions. The sample size m should be the number of users (or the batch size over users). As written, the complexity analysis of O(m^2 d + m^3) in Sec. 4.7.2 is inconsistent with the datasets used, where the number of interactions is on the order of 10^6, and it is unclear what m actually is in the implementation. The authors should clarify the definition and ensure the notation matches the computation.
minor comments (5)
  1. [Sec. 4.4.1, text] The phrase "we approximate the Gumbel softmax trick [25] is used" contains a grammatical error and should be rewritten.
  2. [Eq. (19)] The denominator contains a typographical artifact "1/M ˝M" and the formatting is unclear; the summation should be written with standard notation.
  3. [Sec. 4.4.2, Eq. (14)] The role of d_θ as a "sensitive feature detector" is not justified by any auxiliary supervision; the only training signal comes indirectly through the HSIC loss. The paper should explain what prevents d_θ from collapsing to a trivial constant mask.
  4. [Sec. 5.4, Figure 3] The RBF kernel bandwidth σ in Eq. (16) is not listed among the tuned hyperparameters, and the figure caption says all y-axis values are scaled by 100 but the axis labels are ambiguous. Please clarify the choice of σ and the axis scaling.
  5. [Throughout] There are several typos and formatting issues, including "Gumble-Softmax" in reference [25], the notation "FairDDA-L𝒅𝒍" in Table 3, and the appearance of "GCN!!" in Figure 1. These should be cleaned up.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: FairDDA's augmentation and debiasing objectives are externally evaluated, and the Eq. (10) sign inconsistency is a correctness concern rather than a by-construction reduction.

full rationale

FairDDA does not reduce to its inputs by construction. The fairness gains are measured with DP and EO on final TopK rankings (Eqs. (23)-(24)), and neither DP nor EO appears in the training loss (Eq. (22)); the augmentation probabilities, feature masks, HSIC objective, and contrastive losses are all defined from model representations and are not fitted to the reported fairness metrics. The only self-referential aspect is that the augmented graph is built from the current debiased representations and then used to train those representations, which is a bootstrap/self-supervised loop rather than an identity or an in-sample prediction. The paper's cited prior work by a co-author (FairCoRe, [3]) is used as a baseline, not as load-bearing justification. A genuine correctness issue exists: Eq. (10) as written assigns low pruning probability to edges ranked lower by the debiased model, the opposite of Hypothesis 1, while the text calls p_uv a pruning probability even though Eq. (7) removes edges only when B_uv=0. That sign/notation inconsistency undermines the stated mechanism but does not make the central claim equivalent to its own inputs, so it does not count as circularity under the requested criteria.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim depends on two heuristic definitions of sensitive edges and features, a standard dependence measure (HSIC), and the LightGCN backbone. The three loss weights are tuned on the benchmark datasets, and two important hyperparameters (tau and sigma) are not reported, leaving the configuration under-specified.

free parameters (5)
  • lambda_r = 1 (best)
    Loss weight for the reconstruction loss in Eq. (22), tuned on ML-1M via grid search over {0.1, 0.5, 1, 5, 10}.
  • lambda_c = 0.1 (best)
    Loss weight for the contrastive loss, tuned on ML-1M over {0.01, 0.05, 0.1, 0.5, 1}.
  • lambda_d = 30 (best)
    Loss weight for the HSIC debiasing loss, tuned on ML-1M over {10, 20, 30, 40, 50}.
  • tau (Gumbel temperature) = not reported
    Temperature in Eq. (12) for the approximate Bernoulli sampling; no value or tuning range is given.
  • sigma (RBF kernel bandwidth) = not reported
    Bandwidth in Eq. (16) for the HSIC kernel; no value or tuning range is given.
assumptions (4)
  • domain assumption Hypothesis 1: sensitive edges are those where the debiased model ranks the item lower than the performance-oriented model.
    Stated in Section 4.2 and used in Eq. (10). It is a heuristic, not derived from first principles.
  • domain assumption Hypothesis 2: sensitive features are those where the debiased and biased representations are highly similar at that dimension.
    Stated in Section 4.2 and used in Eq. (14) to generate feature masks.
  • domain assumption Minimizing HSIC between augmented and biased representations reduces sensitive information in the augmented representations.
    Assumed in Section 4.5, following prior HSIC-based debiasing works such as FairIB.
  • domain assumption LightGCN is an appropriate base encoder, and sensitive attributes propagate through graph convolution as described.
    Adopted in Section 3.2, based on prior findings about GCN propagation and homophily.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving Recommendation Fairness via Graph Structure and Representation Augmentation." pith.science (2026). https://pith.science/paper/5HJGP5X6

@misc{pith2026250819547,
  author       = {Pith},
  title        = {Pith review of: Improving Recommendation Fairness via Graph Structure and Representation Augmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5HJGP5X6}},
  note         = {Machine review of arXiv:2508.19547}
}
read the original abstract

Graph Convolutional Networks (GCNs) have become increasingly popular in recommendation systems. However, recent studies have shown that GCN-based models will cause sensitive information to disseminate widely in the graph structure, amplifying data bias and raising fairness concerns. While various fairness methods have been proposed, most of them neglect the impact of biased data on representation learning, which results in limited fairness improvement. Moreover, some studies have focused on constructing fair and balanced data distributions through data augmentation, but these methods significantly reduce utility due to disruption of user preferences. In this paper, we aim to design a fair recommendation method from the perspective of data augmentation to improve fairness while preserving recommendation utility. To achieve fairness-aware data augmentation with minimal disruption to user preferences, we propose two prior hypotheses. The first hypothesis identifies sensitive interactions by comparing outcomes of performance-oriented and fairness-aware recommendations, while the second one focuses on detecting sensitive features by analyzing feature similarities between biased and debiased representations. Then, we propose a dual data augmentation framework for fair recommendation, which includes two data augmentation strategies to generate fair augmented graphs and feature representations. Furthermore, we introduce a debiasing learning method that minimizes the dependence between the learned representations and sensitive information to eliminate bias. Extensive experiments on two real-world datasets demonstrate the superiority of our proposed framework.

Figures

Figures reproduced from arXiv: 2508.19547 by the authors.

Figure 1
Figure 1. An example of sensitive attributions propagation [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The framework of the proposed method, with details shown only for the user side. The key differences from the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Visualization of different methods w.r.t. the gender [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figures from the paper (1 more)
Figure 3
Figure 3. Figure 3: Hyperparameter experiment on ML-1M, with all [PITH_FULL_IMAGE:figures/full_fig_p009_3.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 46 canonical work pages

  1. [1]

    Chirag Agarwal, Himabindu Lakkaraju, and Marinka Zitnik. 2021. Towards a unified framework for fair and stable graph representation learning. In UAI. PMLR, 2114–2124

    work page 2021

  2. [2]

    Yahav Bechavod, Christopher Jung, and Steven Z Wu. 2020. Metric-free individual fairness in online learning. In NeurIPS. 11214–11225

    work page 2020

  3. [3]

    Chenzhong Bin, Wenqiang Liu, Feng Zhang, Liang Chang, and Tianlong Gu. 2025. FairCoRe: Fairness-aware Recommendation through Counterfactual Representa- tion Learning. IEEE Trans. Knowl. Data Eng. (2025)

    work page 2025

  4. [4]

    Avishek Bose and William Hamilton. 2019. Compositional fairness constraints for graph embeddings. In ICML. PMLR, 715–724

    work page 2019

  5. [5]

    Robin Burke, Nasim Sonboli, and Aldo Ordonez-Gauger. 2018. Balanced neigh- borhoods for multi-sided fairness in recommendation. In FAT. PMLR, 202–214

    work page 2018

  6. [6]

    George Casella and Roger Berger. 2024. Statistical inference . Chapman and Hall/CRC

    work page 2024

  7. [7]

    Òscar Celma Herrada et al. 2009. Music recommendation and discovery in the long tail. Universitat Pompeu Fabra

    work page 2009

  8. [8]

    Lei Chen, Le Wu, Kun Zhang, Richang Hong, Defu Lian, Zhiqiang Zhang, Jun Zhou, and Meng Wang. 2023. Improving recommendation fairness via data augmentation. In WWW. 1012–1020

    work page 2023

Show all 53 references
  1. [9]

    Wei Chen, Yiqing Wu, Zhao Zhang, Fuzhen Zhuang, Zhongshi He, Ruobing Xie, and Feng Xia. 2024. FairGap: Fairness-aware recommendation via generating counterfactual graph. ACM Trans. Inf. Syst. 42, 4 (2024), 1–25

    2024

  2. [10]

    Pengyu Cheng, Weituo Hao, Siyang Yuan, Shijing Si, and Lawrence Carin. 2021. FairFil: Contrastive Neural Debiasing Method for Pretrained Text Encoders. In ICLR

    2021

  3. [11]

    Enyan Dai and Suhang Wang. 2021. Say no to the discrimination: Learning fair graph neural networks with limited sensitive attribute information. In WSDM. 680–688

    2021

  4. [12]

    Yashar Deldjoo, Dietmar Jannach, Alejandro Bellogin, Alessandro Difonzo, and Dario Zanzonelli. 2024. Fairness in recommender systems: research landscape and future directions. User Model. User-Adapt. Interact. 34, 1 (2024), 59–108

    2024

  5. [13]

    Moreno D’Incà, Christos Tzelepis, Ioannis Patras, and Nicu Sebe. 2024. Improving fairness using vision-language driven image augmentation. InW ACV. 4695–4704

    2024

  6. [14]

    Mengnan Du, Fan Yang, Na Zou, and Xia Hu. 2020. Fairness in deep learning: A computational perspective. IEEE Intell. Syst. 36, 4 (2020), 25–34

    2020

  7. [15]

    Cynthia Dwork, Moritz Hardt, Toniann Pitassi, Omer Reingold, and Richard Zemel. 2012. Fairness through awareness. In ITCS. 214–226

    2012

  8. [16]

    Michael D Ekstrand, Mucun Tian, Ion Madrazo Azpiazu, Jennifer D Ekstrand, Oghenemaro Anuyah, David McNeill, and Maria Soledad Pera. 2018. All the cool kids, how do they fit in?: Popularity and demographic biases in recommender evaluation and effectiveness. In FAT. PMLR, 172–186

    2018

  9. [17]

    Arthur Gretton, Olivier Bousquet, Alex Smola, and Bernhard Schölkopf. 2005. Measuring statistical dependence with Hilbert-Schmidt norms. In ALT. Springer, 63–77

    2005

  10. [18]

    Asela Gunawardana and Guy Shani. 2009. A Survey of Accuracy Evaluation Metrics of Recommendation Tasks. J. Mach. Learn. Res. 10, 12 (2009)

    2009

  11. [19]

    Dandan Guo, Chaojie Wang, Baoxiang Wang, and Hongyuan Zha. 2022. Learning fair representations via distance correlation minimization. IEEE Trans. Neural Netw. 35, 2 (2022), 2139–2152

    2022

  12. [20]

    Jihun Ham, Daniel D Lee, Sebastian Mika, and Bernhard Schölkopf. 2004. A kernel view of the dimensionality reduction of manifolds. In ICML. 47

    2004

  13. [21]

    Moritz Hardt, Eric Price, and Nati Srebro. 2016. Equality of opportunity in supervised learning. In NeurIPS. 3315–3323

    2016

  14. [22]

    F Maxwell Harper and Joseph A Konstan. 2015. The movielens datasets: History and context. ACM Trans. Interact. Intell. Syst. 5, 4 (2015), 19

    2015

  15. [23]

    Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. Lightgcn: Simplifying and powering graph convolution network for recommendation. In SIGIR. 639–648

    2020

  16. [24]

    Vasileios Iosifidis, Besnik Fetahu, and Eirini Ntoutsi. 2019. Fae: A fairness-aware ensemble framework. In IEEE BigData. IEEE, 1375–1380

    2019

  17. [25]

    Eric Jang, Shixiang Gu, and Ben Poole. 2017. Categorical Reparametrization with Gumble-Softmax. In ICLR

    2017

  18. [26]

    Kalervo Järvelin and Jaana Kekäläinen. 2000. IR evaluation methods for retrieving highly relevant documents.. In SIGIR. ACM, 41–48

    2000

  19. [27]

    O Deniz Kose and Yanning Shen. 2022. Fair node representation learning via adaptive data augmentation. arXiv:2201.08549 (2022)

    2022 arXiv

  20. [28]

    Yunqi Li, Hanxiong Chen, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang. 2021. User-oriented fairness in recommendation. In WWW. 624–632

    2021

  21. [29]

    Yunqi Li, Hanxiong Chen, Shuyuan Xu, Yingqiang Ge, Juntao Tan, Shuchang Liu, and Yongfeng Zhang. 2022. Fairness in recommendation: A survey. arXiv:2205.13619 (2022)

    2022 arXiv

  22. [30]

    Jianhua Lin. 2002. Divergence measures based on the Shannon entropy. IEEE Trans. Inf. Theory 37, 1 (2002), 145–151

    2002

  23. [31]

    Hongyi Ling, Zhimeng Jiang, Youzhi Luo, Shuiwang Ji, and Na Zou. 2023. Learn- ing fair graph representations via automated data augmentations. In ICLR

    2023

  24. [32]

    Jing Ma, Ruocheng Guo, Mengting Wan, Longqi Yang, Aidong Zhang, and Jun- dong Li. 2022. Learning fair node representations with graph counterfactual fairness. In WSDM. 695–703

    2022

  25. [33]

    Wan-Duo Kurt Ma, JP Lewis, and W Bastiaan Kleijn. 2020. The HSIC bottleneck: Deep learning without back-propagation. In AAAI, Vol. 34. 5085–5092

    2020

  26. [34]

    Changdae Oh, Heeji Won, Junhyuk So, Taero Kim, Yewon Kim, Hosik Choi, and Kyungwoo Song. 2022. Learning fair representation via distributional contrastive disentanglement. In SIGKDD. 1295–1305

    2022

  27. [35]

    Dino Pedreschi, Salvatore Ruggieri, and Franco Turini. 2009. Measuring discrimi- nation in socially-sensitive decision records. In SDM. SIAM, 581–592

    2009

  28. [36]

    Tao Qi, Fangzhao Wu, Chuhan Wu, and Yongfeng Huang. 2021. Personalized news recommendation with knowledge-aware interactive matching. In SIGIR. 61–70

    2021

  29. [37]

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

  30. [38]

    Mhd Hasan Sarhan, Nassir Navab, Abouzar Eslami, and Shadi Albarqouni. 2020. Fairness by learning orthogonal disentangled representations. In ECCV. Springer, 746–761

    2020

  31. [39]

    Connor Shorten and Taghi M Khoshgoftaar. 2019. A survey on image data augmentation for deep learning. J. Big Data 6, 1 (2019), 1–48

    2019

  32. [40]

    Connor Shorten, Taghi M Khoshgoftaar, and Borko Furht. 2021. Text data aug- mentation for deep learning. J. Big Data 8, 1 (2021), 101

    2021

  33. [41]

    Laurens Van der Maaten and Geoffrey Hinton. 2008. Visualizing data using t-SNE. J. Mach. Learn. Res. 9, 11 (2008)

    2008

  34. [42]

    Jizhe Wang, Pipei Huang, Huan Zhao, Zhibo Zhang, Binqiang Zhao, and Dik Lun Lee. 2018. Billion-scale commodity embedding for e-commerce recommendation in alibaba. In SIGKDD. 839–848

    2018

  35. [43]

    Yifan Wang, Weizhi Ma, Min Zhang, Yiqun Liu, and Shaoping Ma. 2023. A survey on the fairness of recommender systems. ACM Trans. Inf. Syst. 41, 3 (2023), 1–43

    2023

  36. [44]

    Yu Wang, Yuying Zhao, Yushun Dong, Huiyuan Chen, Jundong Li, and Tyler Derr. 2022. Improving fairness in graph neural networks via mitigating sensitive attribute leakage. In SIGKDD. 1938–1948

    2022

  37. [45]

    Chuhan Wu, Fangzhao Wu, Xiting Wang, Yongfeng Huang, and Xing Xie. 2021. Fairness-aware news recommendation with decomposed adversarial learning. In AAAI, Vol. 35. 4462–4469

    2021

  38. [46]

    Le Wu, Lei Chen, Pengyang Shao, Richang Hong, Xiting Wang, and Meng Wang

  39. [47]

    Junsong Xie, Yonghui Yang, Zihan Wang, and Le Wu. 2024. Learning fair repre- sentations for recommendation via information bottleneck principle. In IJCAI. 2469–2477

    2024

  40. [48]

    Sirui Yao and Bert Huang. 2017. Beyond parity: Fairness objectives for collabora- tive filtering. In NeurIPS. 2921–2930

    2017

  41. [49]

    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. In SIGIR. 1294–1303

    2022

  42. [50]

    Chen Zhao, Le Wu, Pengyang Shao, Kun Zhang, Richang Hong, and Meng Wang

  43. [2009]

    BPR: Bayesian personalized ranking from implicit feedback. In UAI. 452– 461

  44. [2021]

    Learning fair representations for recommendation: A graph-based perspec- tive. In WWW. 2198–2208

  45. [2023]

    In AAAI, Vol

    Fair representation learning for recommendation: a mutual information perspective. In AAAI, Vol. 37. 4911–4919. Improving Recommendation Fairness via Graph Structure and Representation Augmentation CIKM ’25, November 10–14, 2025, Seoul, Republic of Korea. Table 4: Performance ...

    2025

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.