Pith. sign in

REVIEW 5 major objections 7 minor 64 references

BiFair: A Fairness-aware Training Framework for LLM-enhanced Recommender Systems via Bi-level Optimization

T0 review · 5 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read BiFair claims that jointly optimizing LLM-generated item representations and the trainable projector in a bi-level loop mitigates both prior and training unfairness, reporting lower group-utility variance with no accuracy loss on three…

desk verdict Useful empirical observation and a plausible framing, but the core method as written is internally inconsistent: Eq. (10) has a sign error, and Algorithm 1 drops the second-order term that makes the loop bi-level. read the letter →

arxiv 2507.04294 v1 pith:CL34BT73 submitted 2025-07-06 cs.IR

Jiaming Zhang , Yuyuan Li , Yiqun Xu , Li Zhang , Xiaohua Feng , Zhifei Ren , Chaochao Chen This is my paper · ORCID
classification cs.IR
keywords fairness-awarerecommendationLLM-enhancedrecommendersystemsbi-leveloptimizationitem-sidefairnessgrouprepresentationdebiasingmulti-objectiveentropyregularization
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's central claim is that unfairness in LLM-enhanced recommender systems comes from two separate sources—biases baked into the semantic item representations the frozen LLM produces, and biases amplified while the downstream recommendation projector is trained—and that both can be reduced at once by a bi-level training loop. BiFair does this by alternating an inner optimization step that trains the projector under a fairness-aware loss with an outer step that nudges the LLM-generated representations themselves. This matters because the paper's empirical comparison indicates that LLM-enhanced recommenders are already fairer than traditional ID-based ones, yet the worst item groups still receive far less than average utility; fixing only the training stage leaves the representation-level bias untouched. Reported results on three real-world datasets show that BiFair lowers the coefficient of variation of group utilities and raises the utility of the bottom quarter of groups while keeping recommendation accuracy at parity, suggesting the fairness–utility trade-off can be improved rather than accepted.

What carries the argument

The load-bearing machinery is the bi-level formulation: a lower-level problem that trains the projector, and an upper-level problem that updates the LLM-generated semantic representations in the direction that would make the trained projector fairer. Directly differentiating through the lower-level minimizer is too expensive, so the paper applies a single-gradient-step surrogate for $\theta^*$ and a finite-difference approximation of the cross-gradient term, reducing cost from $O(|Z|\times|\theta|)$ to $O(|Z|+|\theta|)$. The second piece is the adaptive fairness-aware loss: group losses form a vector, and an entropy-regularized conditional-gradient procedure chooses a weighted gradient direction that simultaneously descends every group's loss, with the entropy term preventing the update from collapsing onto one group. That loss is used in both levels, so the same fairness signal shapes the projector and the representations.

What would settle it

Run the BiFair training loop on one of the three datasets, but at each outer update fully train the projector instead of taking a single step, and compare the final coefficient of variation and bottom-quarter utility; if the fairness gains shrink or reverse, the one-step shortcut, not the bi-level idea, is doing the work.

Watch

Extended reading notes

Core claim

BiFair's core discovery is that a single fairness-aware loss, wrapped in a two-level optimization, can act on both stages of an LLM-enhanced recommender at once. Writing the semantic representations as $Z$ and the projector parameters as $\theta$, the inner problem solves $\theta^*(Z)=\arg\min_\theta \mathcal{L}(\theta,Z)$ to remove training unfairness, and the outer problem minimizes $\mathcal{L}(\theta^*(Z),Z)$ over $Z$ to remove prior unfairness inherited from the LLM. Because solving the inner problem exactly at every outer step is expensive, the gradient through $\theta^*(Z)$ is approximated by one step of parameter update and a finite-difference estimate of the second-order term, following the standard differentiable-optimization shortcut. The loss itself is an $N$-dimensional vector of group losses whose weights are chosen, by entropy maximization and a conditional-gradient solver, to point in a direction that helps all item groups rather than concentrating on the currently worst one. In the experiments, this combination reduces the coefficient of variation by about 9% for popularity groups and 8.9% for genre groups relative to the strongest baselines, improves the bottom-25% group utility by roughly 6%, and does so without degrading Recall, NDCG, or Hit Rate.

Load-bearing premise

The load-bearing premise is that one quick retraining step of the projector tells us what fully retraining it would do, so the outer loop is really improving fairness rather than a substitute version of it; if that shortcut is wrong, the claimed removal of prior unfairness could disappear.

Editorial extensions

If this is right

  • If BiFair is right, fairness interventions for LLM-enhanced recommenders should be applied in both stages: debiasing the semantic representation space and debiasing training are complementary, and methods that only touch the loss function leave prior unfairness in place.
  • Because the outer loop changes $Z$ rather than the frozen LLM itself, the framework can be dropped onto any existing LLM-enhanced recommender whose pipeline is 'frozen LLM representations followed by a trainable projector,' which the paper demonstrates on two representative architectures with three different base LLMs.
  • The empirical study's finding that LLM-enhanced recommenders are fairer than traditional ID-based recommenders would redirect the fairness debate: the problem is not that adding LLMs creates unfairness, but that a sizable group gap remains and needs targeted correction.
  • The measured improvements—coefficient of variation down roughly 9% for popularity fairness and 8.9% for genre fairness, with bottom-quarter utility up 5–6%—give a concrete benchmark for later fairness methods to beat.

Reading between the lines

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

  • Editorial inference: if $Z$-updates remove representation-level bias, the same bi-level loop could be applied to frozen encoders beyond LLMs, for example vision or audio encoders feeding a downstream model, wherever the upstream representation is suspected of carrying group bias.
  • Editorial inference: a direct way to isolate the contribution of the two-level loop is to run the adaptive fairness loss with only the inner level and only the outer level separately; the paper's ablation compares against separate training, not against the two one-level versions.
  • Editorial inference: the entropy-regularized conditional-gradient weighting is defined without any group-frequency prior, so it could be tested as a standalone loss in ordinary collaborative filtering, where representation-level prior unfairness does not exist, to separate the loss's contribution from the bi-level structure.
  • Editorial inference: because the one-step approximation is the single most fragile step, a natural stress test is to replace it with several inner steps or fully trained inner parameters and measure whether the fairness gains persist; if they vanish, the reported gains depend on that shortcut.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 7 minor

Summary. The paper studies item-side fairness in LLM-enhanced recommender systems. It decomposes unfairness into prior unfairness inherited from LLM-generated representations and training unfairness introduced by the downstream recommendation model, then proposes BiFair, a bi-level optimization framework that alternately updates the LLM-generated representation Z and the trainable projector theta using a fairness-aware loss. The paper also proposes an entropy-based adaptive inter-group weighting mechanism and reports experiments on Movies, Games, and Books showing improved CV and MIN fairness metrics relative to several baselines.

Significance. The topic is timely: fairness in LLM-enhanced recommender systems is under-explored, and the proposed decomposition into prior and training unfairness is a natural and potentially useful conceptual step. The empirical observation that LLM-enhanced RSs can already be fairer than traditional RSs is also informative. However, the manuscript as written does not specify a coherent algorithm that implements the claimed bi-level mechanism, and the derived objective in the adaptive weighting scheme has a sign error that inverts the entropy-maximization goal. Because these issues bear directly on the central claims, the empirical results in Section 4 cannot currently be attributed to the proposed method. No code is provided, which further impedes verification.

major comments (5)
  1. [Section 3.2.4, Algorithm 1] Algorithm 1 line 10 updates Z using only the first-order term ∇_Z L(θ_t, Z_{t−1}), omitting the second-order term and the one-step-unrolled gradient computed in lines 7–9 and in Eqs. (6)–(7). As written, the outer update is ordinary alternating gradient descent on Z, not the bi-level optimization stated in Eq. (4). The claimed mechanism for mitigating prior unfairness is therefore not implemented by the published algorithm.
  2. [Section 3.3, Eq. (10) vs. Appendix A, Eq. (12)] The bracket in Eq. (10), [p_n log p_n − p_n Σ_i p_i log p_i], is the negative of the coefficient appearing in the entropy gradient derived in Eq. (12), which is p_i(Σ_j p_j log p_j − log p_i). Since Eq. (10) is maximized over w, the written objective minimizes d_t(w)^T ∇H_s rather than maximizing it, contradicting the stated goal in Eq. (9) of maximizing the entropy H_s(L(θ)). This sign error inverts the adaptive inter-group balancing mechanism.
  3. [Section 3.3, final paragraph] The paper explicitly states that the fairness-aware weight w is computed only after determining the update directions for both the inner and outer levels. Consequently, w cannot influence the Z update in the outer level, and Algorithm 1 contains no step that uses w to form the θ update either. The adaptive fairness-aware loss is therefore not integrated into the claimed bi-level loop, breaking the definitional link between Eq. (4)'s loss L(θ,Z) and the fairness mechanism.
  4. [Section 3.2.4, Eqs. (5)–(7)] Even setting aside the Algorithm 1 omission, replacing the inner minimizer θ*(Z) with a single gradient step is a DARTS-style approximation for which no convergence or bias analysis is provided and no empirical validation is offered. The paper gives no evidence that minimizing L(θ', Z) over Z reduces the true outer objective L(θ*(Z), Z), which is the load-bearing step for the claimed mitigation of prior unfairness.
  5. [Section 4.4, Table 2] No variance or significance testing is reported. On Movies, BiFair's NDCG and HR are below AlphaRec in both Pop and Genre fairness blocks (e.g., NDCG 0.1820 vs. 0.1831 and HR 0.4305 vs. 0.4338 in the Pop block), so the Section 4.4.2 claim that BiFair 'outperforms or matches' accuracy-focused baselines is not supported by the table as presented. At minimum, error bars or statistical tests are needed before drawing such a conclusion.
minor comments (7)
  1. [Section 4.1] The heading 'Datastes' should be corrected to 'Datasets'.
  2. [Section 1] The word 'LLMss' appears in the first paragraph and should be corrected.
  3. [Section 4.4.3] 'Morevoer' should be 'Moreover'.
  4. [Table 2] Several entries in Table 2 are malformed, such as '0.77', '0.054', and '0.775'; these should be cleaned up.
  5. [Appendix A, Eq. (12)] The derivation in Eq. (12) uses i both as the outer summation index and as the softmax index, and the text interchanges n and i; this makes the derivation hard to follow.
  6. [Figure 1] The caption 'an average improvement of 30.73% in unfairness in recommendation (NDCG@20) (CV@20)' is unclear and should be rewritten.
  7. [Algorithm 1] Algorithm 1 uses η for both inner and outer updates but Eq. (6) introduces a separate step size ξ; the relationship between η and ξ should be clarified.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the fairness gains are measured empirically, the optimization machinery is externally sourced, and the self-citations are non-load-bearing.

full rationale

BiFair's central claims are empirical: it reports CV/MIN on held-out test splits after training with a fairness-aware loss. The fairness-aware loss is a training objective, not a fitted quantity relabeled as a prediction; no Table 2 number is the by-construction value of Eq. (8)-(11). The bi-level gradient approximation in Eq. (5)-(7) is imported from DARTS [27], an external method, and the Frank-Wolfe/MGDA components are standard external optimization tools. The only self-citations ([14], [61]) support dataset grouping and learning-rate schedule, and neither is load-bearing. The paper does contain an internal inconsistency between the derived outer gradient Eq. (6) and Algorithm 1 line 10, which drops the second-order term, but that is an implementation/correctness concern, not a circular reduction of a prediction to its input. Therefore no circular step is identified.

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

The ledger is intentionally modest: no new physical or model entities are proposed. The load-bearing imports are the DARTS-style approximation, the entropy objective, and the conceptual decomposition of unfairness; the unstated hyperparameters are the main reproducibility cost.

free parameters (4)
  • xi (inner-level step size) = not reported
    Introduced in Eq. (5) for the single-step inner approximation; Section 4.3.3 does not give its value, and the outer gradient depends on it.
  • epsilon (finite-difference scalar) = not reported
    Used in Eq. (7) to approximate the second-order term; no value is provided.
  • Frank-Wolfe iterations T = not reported
    Algorithm 2 loops T times; T is not specified in Section 4.3.3.
  • InfoNCE temperature tau = 0.15/0.2 per dataset, 0.1 in Appendix B.4
    Controls the sharpness of the base recommendation loss; values are chosen per dataset following prior work, and the central results depend on this choice.
assumptions (4)
  • ad hoc to paper The single-step gradient approximation is a valid substitute for the true inner-level minimizer.
    Eqs. (5)-(7) in Section 3.2.4 assume this without convergence or bias analysis for the nonconvex recommendation setting.
  • domain assumption Unfairness decomposes additively into prior unfairness in LLM representations and training unfairness in the projector.
    Section 3.2 states this decomposition, but no experiment isolates the two components.
  • domain assumption Maximizing softmax entropy of the group-loss vector improves item-side fairness as measured by CV and MIN.
    Eqs. (9)-(10) link the surrogate objective to the evaluation metric without a formal equivalence proof.
  • standard math Frank-Wolfe on the Gram matrix finds a descent direction satisfying per-group constraints.
    Algorithm 2 cites Jaggi 2013; the method is standard, but convergence for this entropy-regularized objective is not shown.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BiFair: A Fairness-aware Training Framework for LLM-enhanced Recommender Systems via Bi-level Optimization." pith.science (2026). https://pith.science/paper/CL34BT73

@misc{pith2026250704294,
  author       = {Pith},
  title        = {Pith review of: BiFair: A Fairness-aware Training Framework for LLM-enhanced Recommender Systems via Bi-level Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CL34BT73}},
  note         = {Machine review of arXiv:2507.04294}
}
read the original abstract

Large Language Model-enhanced Recommender Systems (LLM-enhanced RSs) have emerged as a powerful approach to improving recommendation quality by leveraging LLMs to generate item representations. Despite these advancements, the integration of LLMs raises severe fairness concerns. Existing studies reveal that LLM-based RSs exhibit greater unfairness than traditional RSs, yet fairness issues in LLM-enhanced RSs remain largely unexplored. In this paper, our empirical study reveals that while LLM-enhanced RSs improve fairness across item groups, a significant fairness gap persists. Further enhancement remains challenging due to the architectural differences and varying sources of unfairness inherent in LLM-enhanced RSs. To bridge this gap, we first decompose unfairness into i) \textit{prior unfairness} in LLM-generated representations and ii) \textit{training unfairness} in recommendation models. Then, we propose BiFair, a bi-level optimization-based fairness-aware training framework designed to mitigate both prior and training unfairness simultaneously. BiFair optimizes two sets of learnable parameters: LLM-generated representations and a trainable projector in the recommendation model, using a two-level nested optimization process. Additionally, we introduce an adaptive inter-group balancing mechanism, leveraging multi-objective optimization principles to dynamically balance fairness across item groups. Extensive experiments on three real-world datasets demonstrate that BiFair significantly mitigates unfairness and outperforms previous state-of-the-art methods.

Figures

Figures reproduced from arXiv: 2507.04294 by the authors.

Figure 1
Figure 1. Average performance comparison of 5 traditional [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. An illustration of the pipeline of our proposed [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Ablation study of bi-level optimization framework [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

64 extracted references · 39 canonical work pages

  1. [1]

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Floren- cia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774 (2023)

    arXiv 2023

  2. [2]

    Xuechunzi Bai, Angelina Wang, Ilia Sucholutsky, and Thomas L Griffiths. 2024. Measuring Implicit Bias in Explicitly Unbiased Large Language Models. In NeurIPS 2024 Workshop on Behavioral Machine Learning

    work page 2024

  3. [3]

    Xuechunzi Bai, Angelina Wang, Ilia Sucholutsky, and Thomas L Griffiths. 2025. Explicitly unbiased large language models still form biased associations. Pro- ceedings of the National Academy of Sciences 122, 8 (2025), e2416228122

    work page 2025

  4. [4]

    Zhuoxi Bai, Ning Wu, Fengyu Cai, Xinyi Zhu, and Yun Xiong. 2024. Aligning Large Language Model with Direct Multi-Preference Optimization for Recom- mendation. In Proceedings of the 33rd CIKM . 76–86

    work page 2024

  5. [5]

    Keqin Bao, Jizhi Zhang, Wenjie Wang, Yang Zhang, Zhengyi Yang, Yancheng Luo, Chong Chen, Fuli Feng, and Qi Tian. 2023. A bi-step grounding paradigm for large language models in recommendation systems. arXiv preprint arXiv:2308.08434 (2023)

    arXiv 2023

  6. [6]

    Youjun Bao and Xiaohong Jiang. 2016. An intelligent medicine recommender system framework. In 2016 IEEE 11Th conference on industrial electronics and applications (ICIEA). IEEE, 1383–1388

    work page 2016

  7. [7]

    Alex Beutel, Jilin Chen, Tulsee Doshi, Hai Qian, Li Wei, Yi Wu, Lukasz Heldt, Zhe Zhao, Lichan Hong, Ed H Chi, et al . 2019. Fairness in recommendation ranking through pairwise comparisons. In Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining . 2212–2220

    work page 2019

  8. [8]

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

    work page 2018

Show all 64 references
  1. [9]

    Xiao Chen, Wenqi Fan, Jingfan Chen, Haochen Liu, Zitao Liu, Zhaoxiang Zhang, and Qing Li. 2023. Fairly adaptive negative sampling for recommendations. In Proceedings of the ACM Web Conference 2023 . 3723–3733

    2023

  2. [10]

    Jean-Antoine Désidéri. 2012. Multiple-gradient descent algorithm (MGDA) for multiobjective optimization. Comptes Rendus Mathematique 350, 5-6 (2012), 313–318

    2012

  3. [11]

    Cynthia Dwork, Moritz Hardt, Toniann Pitassi, Omer Reingold, and Richard Zemel. 2012. Fairness through awareness. In Proceedings of the 3rd innovations in theoretical computer science conference . 214–226

    2012

  4. [12]

    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 recom- mender evaluation and effectiveness. In Conference on f...

    2018

  5. [13]

    Chongming Gao, Ruijun Chen, Shuai Yuan, Kexin Huang, Yuanqing Yu, and Xi- angnan He. 2024. SPRec: Leveraging Self-Play to Debias Preference Alignment for Large Language Model-based Recommendations. arXiv preprint arXiv:2412.09243 (2024)

    2024 arXiv

  6. [14]

    Zhongxuan Han, Chaochao Chen, Xiaolin Zheng, Weiming Liu, Jun Wang, Wen- jie Cheng, and Yuyuan Li. 2023. In-processing user constrained dominant sets for user-oriented fairness in recommender systems. In Proceedings of the 31st ACM International Conference on Multimedia . 6190–6201

    2023

  7. [15]

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

    2020

  8. [16]

    Yifan Hu, Yehuda Koren, and Chris Volinsky. 2008. Collaborative filtering for implicit feedback datasets. In 2008 Eighth IEEE international conference on data mining. Ieee, 263–272

    2008

  9. [17]

    Martin Jaggi. 2013. Revisiting Frank-Wolfe: Projection-free sparse convex opti- mization. In International conference on machine learning . PMLR, 427–435

    2013

  10. [18]

    Meng Jiang, Keqin Bao, Jizhi Zhang, Wenjie Wang, Zhengyi Yang, Fuli Feng, and Xiangnan He. 2024. Item-side Fairness of Large Language Model-based Recommendation System. In Proceedings of the ACM on Web Conference 2024 . 4717–4726

    2024

  11. [19]

    Yehuda Koren, Robert Bell, and Chris Volinsky. 2009. Matrix factorization tech- niques for recommender systems. Computer 42, 8 (2009), 30–37

    2009

  12. [20]

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

    2020

  13. [21]

    Roger Zhe Li, Julián Urbano, and Alan Hanjalic. 2021. Leave no user behind: Towards improving the utility of recommender systems for non-mainstream users. In Proceedings of the 14th ACM International Conference on Web Search and Data Mining. 103–111

    2021

  14. [22]

    Yunqi Li, Hanxiong Chen, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang. 2021. User-oriented fairness in recommendation. In Proceedings of the web conference

    2021

  15. [23]

    Yunqi Li, Hanxiong Chen, Shuyuan Xu, Yingqiang Ge, Juntao Tan, Shuchang Liu, and Yongfeng Zhang. 2023. Fairness in recommendation: Foundations, methods, and applications. ACM Transactions on Intelligent Systems and Technology 14, 5 (2023), 1–48

    2023

  16. [24]

    Dawen Liang, Rahul G Krishnan, Matthew D Hoffman, and Tony Jebara. 2018. Variational autoencoders for collaborative filtering. In Proceedings of the 2018 world wide web conference . 689–698

    2018

  17. [25]

    Jiayi Liao, Xiangnan He, Ruobing Xie, Jiancan Wu, Yancheng Yuan, Xingwu Sun, Zhanhui Kang, and Xiang Wang. 2024. RosePO: Aligning LLM-based Recommenders with Human Values. arXiv preprint arXiv:2410.12519 (2024)

    2024 arXiv

  18. [26]

    Jiayi Liao, Sihang Li, Zhengyi Yang, Jiancan Wu, Yancheng Yuan, Xiang Wang, and Xiangnan He. 2024. Llara: Large language-recommendation assistant. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval . 1785–1795

    2024

  19. [27]

    Hanxiao Liu, Karen Simonyan, and Yiming Yang. 2019. DARTS: Differentiable Architecture Search. In International Conference on Learning Representations

    2019

  20. [28]

    Liyang Liu, Yi Li, Zhanghui Kuang, Jing-Hao Xue, Yimin Chen, Wenming Yang, Qingmin Liao, and Wayne Zhang. 2021. Towards Impartial Multi-task Learning. In International Conference on Learning Representations

    2021

  21. [29]

    Hanjia Lyu, Song Jiang, Hanqing Zeng, Yinglong Xia, Qifan Wang, Si Zhang, Ren Chen, Christopher Leung, Jiajie Tang, and Jiebo Luo. 2023. Llm-rec: Person- alized recommendation via prompting large language models. arXiv preprint arXiv:2307.15780 (2023)

    2023 arXiv

  22. [30]

    Rui Meng, Ye Liu, Shafiq Rayhan Joty, Caiming Xiong, Yingbo Zhou, and Semih Yavuz. 2024. Sfrembedding-mistral: enhance text retrieval with transfer learning. Salesforce AI Research Blog 3 (2024), 6

    2024

  23. [31]

    Aaron van den Oord, Yazhe Li, and Oriol Vinyals. 2018. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748 (2018)

    2018 arXiv

  24. [32]

    Ioannis Paparrizos, B Barla Cambazoglu, and Aristides Gionis. 2011. Machine learned job recommendation. In Proceedings of the fifth ACM Conference on Recommender Systems. 325–328

    2011

  25. [33]

    Gourab K Patro, Lorenzo Porcaro, Laura Mitchell, Qiuyue Zhang, Meike Zehlike, and Nikhil Garg. 2022. Fair ranking: a critical review, challenges, and future directions. In Proceedings of the 2022 ACM conference on fairness, accountability, and transparency. 1929–1942

    2022

  26. [34]

    Bashir Rastegarpanah, Krishna P Gummadi, and Mark Crovella. 2019. Fighting fire with fire: Using antidote data to improve polarization and fairness of recom- mender systems. In Proceedings of the twelfth ACM international conference on web search and data mining . 231–239

    2019

  27. [35]

    Xubin Ren, Wei Wei, Lianghao Xia, Lixin Su, Suqi Cheng, Junfeng Wang, Dawei Yin, and Chao Huang. 2024. Representation learning with large language models for recommendation. In Proceedings of the ACM Web Conference 2024. 3464–3475

    2024

  28. [36]

    Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt- Thieme. 2009. BPR: Bayesian personalized ranking from implicit feedback. In Proceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence . 452–461

    2009

  29. [37]

    Shiori Sagawa, Pang Wei Koh, Tatsunori B Hashimoto, and Percy Liang. 2019. Distributionally Robust Neural Networks. InInternational Conference on Learning Representations

    2019

  30. [38]

    Leheng Sheng, An Zhang, Yi Zhang, Yuxin Chen, Xiang Wang, and Tat-Seng Chua. 2024. Language Representations Can be What Recommenders Need: Jiaming Zhang, Yuyuan Li, Yiqun Xu, Li Zhang, Xiaohua Feng, Zhifei Ren, and Chaochao Chen Findings and Potentials. arXiv preprint arXiv:24...

    2024 arXiv

  31. [39]

    Maria Stratigi, Jyrki Nummenmaa, Evaggelia Pitoura, and Kostas Stefanidis. 2020. Fair sequential group recommendations. In Proceedings of the 35th Annual ACM Symposium on Applied Computing. 1443–1452

    2020

  32. [40]

    Antonela Tommasel. 2024. Fairness Matters: A look at LLM-generated group recommendations. In Proceedings of the 18th ACM Conference on Recommender Systems. 993–998

    2024

  33. [41]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. 2023. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971 (2023)

    2023 arXiv

  34. [42]

    Jason Turcotte, Chance York, Jacob Irving, Rosanne M Scholl, and Raymond J Pingree. 2015. News recommendations from social media opinion leaders: Ef- fects on media trust and information seeking. Journal of computer-mediated communication 20, 5 (2015), 520–535

    2015

  35. [43]

    Jiayin Wang, Weizhi Ma, Jiayu Li, Hongyu Lu, Min Zhang, Biao Li, Yiqun Liu, Peng Jiang, and Shaoping Ma. 2022. Make fairness more fair: Fair item utility estimation and exposure re-distribution. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data M...

    2022

  36. [44]

    Xiuling Wang and Wendy Hui Wang. 2022. Providing item-side individual fair- ness for deep recommender systems. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency. 117–127

    2022

  37. [45]

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

    2023

  38. [46]

    Yifan Wang, Peijie Sun, Weizhi Ma, Min Zhang, Yuan Zhang, Peng Jiang, and Shaoping Ma. 2024. Intersectional two-sided fairness in recommendation. In Proceedings of the ACM Web Conference 2024 . 3609–3620

    2024

  39. [47]

    Wei Wei, Xubin Ren, Jiabin Tang, Qinyong Wang, Lixin Su, Suqi Cheng, Junfeng Wang, Dawei Yin, and Chao Huang. 2024. Llmrec: Large language models with graph augmentation for recommendation. In Proceedings of the 17th ACM International Conference on Web Search and Data Mining . 806–815

    2024

  40. [48]

    Hongyi Wen, Xinyang Yi, Tiansheng Yao, Jiaxi Tang, Lichan Hong, and Ed H Chi. 2022. Distributionally-robust recommendations for improving worst-case user experience. In Proceedings of the ACM Web Conference 2022 . 3606–3610

    2022

  41. [49]

    Haolun Wu, Chen Ma, Bhaskar Mitra, Fernando Diaz, and Xue Liu. 2022. A multi-objective optimization framework for multi-stakeholder fairness-aware recommendation. ACM Transactions on Information Systems 41, 2 (2022), 1–29

    2022

  42. [50]

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

    2021

  43. [51]

    Yunjia Xi, Weiwen Liu, Jianghao Lin, Xiaoling Cai, Hong Zhu, Jieming Zhu, Bo Chen, Ruiming Tang, Weinan Zhang, and Yong Yu. 2024. Towards open-world recommendation with knowledge augmentation from large language models. In Proceedings of the 18th ACM Conference on Recommender ...

    2024

  44. [52]

    An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, et al . 2024. Qwen2. 5 technical report. arXiv preprint arXiv:2412.15115 (2024)

    2024 arXiv

  45. [53]

    Hao Yang, Zhining Liu, Zeyu Zhang, Chenyi Zhuang, and Xu Chen. 2023. To- wards robust fairness-aware recommendation. In Proceedings of the 17th ACM Conference on Recommender Systems . 211–222

    2023

  46. [54]

    Kai-Ching Yeh, Jou-An Chi, Da-Chen Lian, and Shu-Kai Hsieh. 2023. Evaluat- ing interfaced llm bias. In Proceedings of the 35th Conference on Computational Linguistics and Speech Processing (ROCLING 2023) . 292–299

    2023

  47. [55]

    Junliang Yu, Xin Xia, Tong Chen, Lizhen Cui, Nguyen Quoc Viet Hung, and Hongzhi Yin. 2023. XSimGCL: Towards extremely simple graph contrastive learn- ing for recommendation. IEEE Transactions on Knowledge and Data Engineering 36, 2 (2023), 913–926

    2023

  48. [56]

    Meike Zehlike, Ke Yang, and Julia Stoyanovich. 2022. Fairness in ranking, part ii: Learning-to-rank and recommender systems. Comput. Surveys 55, 6 (2022), 1–41

    2022

  49. [57]

    An Zhang, Yuxin Chen, Leheng Sheng, Xiang Wang, and Tat-Seng Chua. 2024. On generative agents in recommendation. In Proceedings of the 47th international ACM SIGIR conference on research and development in Information Retrieval. 1807– 1817

    2024

  50. [58]

    Jizhi Zhang, Keqin Bao, Yang Zhang, Wenjie Wang, Fuli Feng, and Xiangnan He. 2023. Is chatgpt fair for recommendation? evaluating fairness in large language model recommendation. In Proceedings of the 17th ACM Conference on Recommender Systems. 993–999

    2023

  51. [59]

    Junjie Zhang, Ruobing Xie, Yupeng Hou, Xin Zhao, Leyu Lin, and Ji-Rong Wen

  52. [60]

    Kaidong Zhang and Dong Liu. 2023. Customized segment anything model for medical image segmentation. arXiv preprint arXiv:2304.13785 (2023)

    2023 arXiv

  53. [61]

    Li Zhang, Youwei Liang, Ruiyi Zhang, Amirhosein Javadi, and Pengtao Xie. 2024. BLO-SAM: bi-level optimization based finetuning of the segment anything model for overfitting-preventing semantic segmentation. In Forty-first International Conference on Machine Learning

    2024

  54. [62]

    Jujia Zhao, Wenjie Wang, Xinyu Lin, Leigang Qu, Jizhi Zhang, and Tat-Seng Chua

  55. [64]

    In Proceedings of the 32nd ACM International Conference on Information and Knowledge Management

    Popularity-aware distributionally robust optimization for recommendation system. In Proceedings of the 32nd ACM International Conference on Information and Knowledge Management. 4967–4973. BiFair: A Fairness-aware Training Framework for LLM-enhanced Recommender Systems via Bi-...

    1969

  56. [2023]

    ACM Transactions on Information Systems (2023)

    Recommendation as instruction following: A large language model em- powered recommendation approach. ACM Transactions on Information Systems (2023)

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.