Pith. sign in

REVIEW 5 major objections 6 minor 1 cited by

Bias Mitigation Agent: Optimizing Source Selection for Fair and Balanced Knowledge Retrieval

T0 review · 5 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A source-selection gate powered by a bias classifier cuts biased retrieval outputs by 81.82% over a relevance-only baseline while preserving relevance, in the paper's best configuration.

desk verdict The headline claim is not supported: the same classifier filters sources and scores the outcome, so the 81.82% bias reduction is largely circular. read the letter →

arxiv 2508.18724 v1 pith:MAOK5G4V submitted 2025-08-26 cs.AI cs.CL

Karanbir Singh , Deepak Muppiri , William Ngu This is my paper
classification cs.AIcs.CL
keywords InformationRetrievalAgentsAugmentedGenerationLargeLanguageModelsBiasFairnessmulti-agentsystemsourceselection
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 sets out to show that bias in knowledge retrieval can be cut at the point of source selection, before a language model writes an answer. It builds a supervisor-based multi-agent system in which a knowledge agent retrieves candidate documents, a bias-detection agent scores each candidate using Dbias, a source-selector agent passes only high-confidence unbiased documents to the writer, and the manager enforces retries when no candidate passes the gate. On 112 queries drawn from two annotated news datasets, the zero-shot selector with GPT-4o-mini lowered the bias rate from 49.11% to 8.93%, an 81.82% reduction relative to a naive relevance-only retrieval baseline, while average relevance rose. The authors treat this as evidence that agentic source selection can make retrieval fairer without sacrificing relevance.

What carries the argument

The load-bearing object is the Source Selection Agent's threshold gate: only candidates with binary bias label γ=0 and detector confidence β≥0.7 are eligible, and among those the highest-relevance document wins. The Manager Agent powers retries by query expansion when no candidate passes, while few-shot mode replaces the raw threshold with example-conditioned scoring. This gate is the entire mechanism that separates the new workflow from ordinary relevance-based retrieval.

What would settle it

Have independent human annotators label a held-out sample of the generated answers from baseline and zero-shot modes without knowing the mode; if the human-labeled bias-rate gap is far below 81.82%, the reported reduction is an artifact of Dbias. Also test for label instability by rewording the same query and checking whether the same document's Dbias label flips, which would show the retry loop can be gamed.

Watch

Extended reading notes

Core claim

The central claim is that a bias filter inserted between retrieval and generation reduces biased output by roughly 82% compared with a naive retrieval strategy, and that the reduction does not come at the cost of relevance. The mechanism is a threshold rule: accept only candidate documents with binary bias label γ=0 and detector confidence β≥0.7, then choose the most relevant of those; if none qualify, expand the query and retry. In few-shot mode, the same decision is guided by in-context examples rather than a hard threshold. The best configuration (zero-shot, GPT-4o-mini) achieved 8.93% biased outputs versus 49.11% for the baseline, and its average relevance of 0.366 exceeded the baseline'

Load-bearing premise

The claimed reduction depends on Dbias's labels being the right measure of bias, because the selector rejects anything Dbias flags and the final bias rate counts Dbias labels on the outputs.

Editorial extensions

If this is right

  • In the paper's best configuration, bias-gated selection lowered biased outputs from 49.11% to 8.93% while average relevance rose from 0.169 to 0.366.
  • The mechanism is model- and mode-dependent: GPT-4.1 and GPT-4.1-mini still improve, but by smaller margins, and few-shot is usually less aggressive than zero-shot on bias for GPT-4o-mini.
  • Because the writer is told to rely only on the selected source, the fairness gain comes from input choice rather than output rewriting, keeping the intervention transparent and modular.
  • Few-shot selection retries far less often than zero-shot, so a substantial fairness gain can be obtained at lower latency in that mode.

Reading between the lines

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

  • The measured reduction may be inflated by measurement coupling: the selector filters on Dbias labels and the reported bias rate is computed from the same labels, so part of the gain could be detector-error filtering rather than reader-visible fairness.
  • The retry mechanism invites gaming: each retry expands the query, so the system could keep rephrasing until a document happens to be classified unbiased, selecting for clean-looking rather than genuinely balanced sources.
  • Choosing exactly one source means opposing viewpoints cannot be combined; an extension that lets the writer synthesize several complementary sources could improve balance without relaxing the bias gate.
  • The architecture is portable only where a trustworthy bias detector exists; with an unreliable detector, the gate would give false confidence in downstream fairness.
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 / 6 minor

Summary. The paper introduces the Bias Mitigation Agent, a multi-agent framework for retrieval-augmented generation that aims to reduce bias in retrieved knowledge. A manager agent coordinates a knowledge agent (retrieval via ChromaDB), a bias detection agent (using the Dbias classifier to assign bias confidence β and binary label γ), a source selection agent (zero-shot rule-based filtering or few-shot example-conditioned selection), and a writer agent. Three modes are compared: no source selection (baseline, top-1 retrieval), zero-shot selection, and few-shot selection, across three GPT models on 112 curated queries from MBIC/BABE. The headline result is an 81.82% reduction in bias rate for GPT-4o-mini in zero-shot mode compared to baseline (8.93% vs. 49.11%), with smaller reductions reported for other model/mode combinations.

Significance. If the reported effectiveness were established, the paper would offer a practical, modular architecture for fairness-aware source selection in RAG/agentic systems, with useful engineering components: explicit state management, retry logic, and a comparison of zero-shot and few-shot selectors. The use of MBIC/BABE news datasets is appropriate for the problem domain. However, the central empirical claim is not currently supported because the evaluation metric is the same classifier used to select sources, and the baseline is confounded with candidate-pool size. The paper's potential is real, but the evidence as presented does not demonstrate a genuine fairness improvement.

major comments (5)
  1. [§3.3.2, §3.3.3, Eq. (4), §4.1, Fig. 5] The evaluation is circular. The Dbias classifier in §3.3.2 produces the binary label γ and confidence β used in Eq. (4) to select only candidates with γ=0 and β≥0.7 (and analogously for few-shot). The outcome metric in §4.1 and Fig. 5 is the fraction of outputs the same Dbias-style classifier labels biased. Filtering on a classifier's own labels mechanically lowers its positive rate, even if the classifier is noisy or unrelated to reader-relevant bias. The reported 81.82% reduction is therefore not evidence of genuine bias mitigation. An independent outcome measure is required: e.g., scoring final outputs with human annotations from MBIC/BABE, or a second, independently trained bias classifier, and reporting agreement/error rates. Without such an external signal, the central claim is unsubstantiated.
  2. [§3.3.1, §3.3.3, Eqs. (3)–(4)] The baseline and the proposed agent differ in two confounded ways. The baseline retrieves a single top-1 document by Eq. (3), while the agent retrieves a candidate set C and then selects from it in Eq. (4). Thus the comparison conflates candidate-pool size with bias-based selection. To isolate the effect of source selection, the baseline should be run on the same candidate pool (e.g., randomly selecting among top-k candidates, or using top-1 from the same retrieved set), and the results reported separately for first-attempt and final outputs.
  3. [§3.3.1, Fig. 6, §4.2] The retry/query-expansion mechanism is a second confound. When no candidate meets the selection criteria, the knowledge agent rewrites the query based on the rejection reason ρ and retrieves a new candidate set. The retry rate is high (70.54% for GPT-4o-mini zero-shot in Fig. 6), and the final relaxed selection still relies on Dbias scores. The paper does not report how much of the bias reduction is due to the selection rule versus query rewriting, nor does it analyze first-attempt vs. final-attempt outcomes. Such an analysis is needed to attribute the improvement to the source selection mechanism as claimed.
  4. [§4.1, Fig. 5, Table 1] The experimental report lacks statistical grounding. Results are based on 112 queries, but no confidence intervals, standard errors, or significance tests are given for the bias rates shown in Fig. 5. The headline 81.82% reduction is for one model/mode; other rows show smaller gains (e.g., GPT-4.1-mini zero-shot drops from 52.68% to 27.68%). The paper should provide uncertainty estimates and, ideally, paired comparisons across models and modes to support claims that the agent 'significantly outperforms' the baseline.
  5. [§4.2, §4.3, Table 1] There are internal inconsistencies in the reported relevance scores and in the text. Eq. (2) defines ρ ∈ [0,1], but Table 1 lists negative relevance values (e.g., -0.058). The text in §4.2 states that GPT-4o-mini zero-shot's 'beat average relevance score was 0.366... even better than the baseline,' yet Table 1 shows 4o-mini zero-shot relevance of 0.157 ± 0.078 versus 0.169 ± 0.092 for the baseline. Clarify the relevance metric, its range, and correct the comparative claims.
minor comments (6)
  1. [Abstract/Introduction] The phrase 'Now a days' in §1 is informal; should be 'Nowadays'.
  2. [§4.2] The text says 'The beat average relevance score' — likely a typo for 'The best average relevance score.'
  3. [§4.3] The sentence 'it achieved a bias rate of 14.3%, demonstrating a substantial improvement over the baseline mode by 69.48%' is ambiguous: state the absolute rates and the relative reduction clearly.
  4. [§4.1, Fig. 5] Figure 5 lacks error bars or uncertainty intervals; add them or indicate the per-query variance. Also define what 'bias rate' is precisely (fraction of outputs labeled biased by Dbias?) in the caption or text.
  5. [§3.3.3] Eq. (5) defines f_few-shot as 'learned or example-conditioned scoring function implicitly encoded via prompt demonstrations,' but no details of the few-shot examples (size, content, selection) are given. Include them or a reference to a supplemental appendix for reproducibility.
  6. [General] The paper does not include a limitations section. Given the evaluation issues raised above, a candid discussion of the reliance on Dbias and the lack of human evaluation is necessary.

Circularity Check

1 steps flagged · score 8.0 of 10

Dbias is both the selection filter and the outcome metric, so the claimed 81.82% bias reduction is largely forced by construction.

  1. self definitional [Section 3.3.2 (Bias Detection Agent) and Eq. 4 in Section 3.3.3; outcome in Section 4.1 / Figure 5]
    "Each candidate document c_i ∈ C is then analyzed using a pre-trained text classification model called Dbias [16]. ... A binary label (γ_i ∈ {0, 1}), where γ_i = 1 indicates that the document is biased, and γ_i = 0 indicates that it is unbiased. ... α = arg max_{c_i ∈ C′} ρ_i, where C′ = {c_i ∈ C | γ_i = 0 ∧ β_i ≥ 0.7} ... with 49.11% of the outputs labeled biased using GPT-4o-mini"

    The source-selection rule (Eq. 4) admits only documents that Dbias labels unbiased (γ=0, high confidence). The headline outcome, 'bias rate' in Figure 5, is the fraction of outputs labeled biased by the same Dbias classifier (Section 4.1). Filtering on γ=0 mechanically lowers the classifier's positive rate among selected sources, so the reported drop from 49.11% to 8.93% (an 81.82% reduction) is an accounting consequence of the selection rule rather than an independently measured fairness improvement. Few-shot (Eq. 5) uses the same (β, γ, ρ) quantities and inherits the same circularity. No external outcome label (MBIC/BABE human annotations, an independent bias classifier, or human evaluation) is reported for the final outputs. Relevance is measured independently, so relevance claims are n

full rationale

The central empirical claim of an 81.82% bias reduction is not self-contained against an external benchmark: the same Dbias classifier both drives selection (β, γ in Section 3.3.2 and Eq. 4) and defines the evaluation metric ('bias rate', the fraction of outputs labeled biased, Section 4.1/Figure 5). Selecting documents with γ=0 guarantees the selected source is classified unbiased, so the classifier's positive rate on outputs derived from those sources is reduced by construction. The paper mentions MBIC and BABE human-annotated datasets, but does not use them to score the final answers; no independent bias measure is reported. The few-shot mode is equally affected because it conditions on the same (β, γ, ρ) values. Self-citation [19] appears in related work but is not load-bearing. Relevance and latency comparisons are independent and not circular; however, the headline bias-reduction claim reduces to matching a filter to its own evaluation signal, warranting a high circularity score.

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

No new physical or formal entities are posited; the agents are software roles. The quantitative claim rests almost entirely on domain assumptions about Dbias's validity and the test setup, with only one explicitly stated numerical threshold (0.7) and several unspecified system parameters.

free parameters (5)
  • Bias confidence threshold in zero-shot selection = 0.7
    Equation 4 requires beta_i >= 0.7; this threshold controls how many candidates survive and is not derived or justified.
  • Top-k retrieval count
    The Knowledge Agent retrieves an unspecified number of candidate documents; the amount of selection possible depends on this value.
  • Few-shot demonstration set
    Few-shot selection is defined only as labeled examples encoding decision patterns; the examples are not shown, so their content is an implicit parameter.
  • Maximum retries and relaxed selection rules
    The state includes mu and retry logic, and the final attempt applies relaxed selection rules, but no values or rules are specified.
  • Query expansion method on retry
    On rejection the manager transforms q into q' based on the rejection reason; the transformation is unspecified although it changes the candidate pool.
assumptions (5)
  • domain assumption Dbias's bias labels and confidence scores are a valid ground-truth measure of bias for news articles and final LLM answers.
    All selection and evaluation use Dbias in Sections 3.3.2, 4.1, and 4.2. If Dbias is wrong or gameable, the reported bias reduction is not real.
  • domain assumption The MBIC/BABE corpora plus 112 curated queries are representative enough to assess general system behavior.
    Section 4 states curated queries but gives no selection protocol or coverage analysis; external validity is assumed.
  • domain assumption The LLM reasoners faithfully execute the manager and worker roles and do not introduce uncontrolled biases.
    The system's behavior depends on prompt-following in Sections 3.2 and 3.3; no checks for instruction adherence or agent failures are reported.
  • domain assumption The retrieval corpus contains enough genuinely unbiased relevant documents for selection to matter.
    No analysis of candidate pool composition is given; negative relevance scores and high retry rates suggest candidates can be poor.
  • domain assumption The relevance score rho from vector similarity is comparable across modes and faithfully reflects answer quality.
    Table 1 uses relevance as utility, but no human relevance evaluation is provided; negative minimum values and high variance are not explained.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bias Mitigation Agent: Optimizing Source Selection for Fair and Balanced Knowledge Retrieval." pith.science (2026). https://pith.science/paper/MAOK5G4V

@misc{pith2026250818724,
  author       = {Pith},
  title        = {Pith review of: Bias Mitigation Agent: Optimizing Source Selection for Fair and Balanced Knowledge Retrieval},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MAOK5G4V}},
  note         = {Machine review of arXiv:2508.18724}
}
read the original abstract

Large Language Models (LLMs) have transformed the field of artificial intelligence by unlocking the era of generative applications. Built on top of generative AI capabilities, Agentic AI represents a major shift toward autonomous, goal-driven systems that can reason, retrieve, and act. However, they also inherit the bias present in both internal and external information sources. This significantly affects the fairness and balance of retrieved information, and hence reduces user trust. To address this critical challenge, we introduce a novel Bias Mitigation Agent, a multi-agent system designed to orchestrate the workflow of bias mitigation through specialized agents that optimize the selection of sources to ensure that the retrieved content is both highly relevant and minimally biased to promote fair and balanced knowledge dissemination. The experimental results demonstrate an 81.82\% reduction in bias compared to a baseline naive retrieval strategy.

Figures

Figures reproduced from arXiv: 2508.18724 by the authors.

Figure 1
Figure 1. Architecture of a Supervisor-Based Multi-Agent System for Bias Mitigation. The diagram depicts the life cycle of the [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. System prompt used for manager when source se [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 2
Figure 2. System prompt used for manager when source se [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Execution flow of the Bias Mitigation Framework [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Bias rate comparison for each source selection [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Retry rates for Zero-Shot and Few-Shot selectors [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Aligned Agents, Biased Swarm: Measuring Bias Amplification in Multi-Agent Systems

    cs.MA 2026-04 unverdicted novelty 6.0 of 10

    Multi-agent systems amplify minor stochastic biases into systemic polarization via echo-chamber effects in structured workflows, even with neutral agents.

Reference graph

Works this paper leans on

26 extracted references · 13 canonical work pages · cited by 1 Pith paper

  1. [19]

    Karanbir Singh and William Ngu. 2025. Bias-Aware Agent: Enhancing Fairness in AI-Driven Knowledge Retrieval. https://arxiv.org/abs/2503.21237

    work page Pith review arXiv 2025

  2. [1]

    Tolga Bolukbasi, Kai-Wei Chang, James Zou, Venkatesh Saligrama, and Adam Kalai. 2016. Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings. arXiv (2016). arXiv:1607.06520 [cs.CL] https: //arxiv.org/abs/1607.06520

    arXiv 2016

  3. [2]

    Angana Borah and Rada Mihalcea. 2024. Towards Implicit Bias Detection and Mitigation in Multi-Agent LLM Interactions. arXiv:2410.02584 [cs.CL] https: //arxiv.org/abs/2410.02584

    arXiv 2024

  4. [3]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwin...

    arXiv 2020

  5. [4]

    Chroma. 2022. Chroma: The open-source AI application database. https://www. trychroma.com/ Accessed: May 2025

    work page 2022

  6. [5]

    Maria De-Arteaga, Alexey Romanov, Hanna Wallach, Jennifer Chayes, Chris- tian Borgs, Alexandra Chouldechova, Sahin Geyik, Krishnaram Kenthapadi, and Adam Tauman Kalai. 2019. Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting. InProceedings of the Conference on Fairness, Account- ability, and Transparency (FAT* ’19). ACM, 1...

    arXiv 2019

  7. [6]

    Yufei Guo, Muzhe Guo, Juntao Su, Zhou Yang, Mengqiu Zhu, Hongfei Li, Mengyang Qiu, and Shuo Shuo Liu. 2024. Bias in Large Language Models: Origin, Evaluation, and Mitigation. arXiv:2411.10915 [cs.CL] https://arxiv.org/ abs/2411.10915

    arXiv 2024

  8. [7]

    Mengxuan Hu, Hongyi Wu, Zihan Guan, Ronghang Zhu, Dongliang Guo, Daiqing Qi, and Sheng Li. 2024. No Free Lunch: Retrieval-Augmented Generation Un- dermines Fairness in LLMs, Even for Vigilant Users. arXiv:2410.07589 [cs.IR] https://arxiv.org/abs/2410.07589

    arXiv 2024

Show all 26 references
  1. [8]

    LangChain Inc. 2023. LangGraph: A Library for Building Multi-Agent Workflows with LLMs. https://github.com/langchain-ai/langgraph Accessed: May 2025

    2023

  2. [9]

    Thomas Jaenich, Graham McDonald, and Iadh Ounis. 2024. Fairness-Aware Exposure Allocation via Adaptive Reranking. In Proceedings of the 47th Inter- national ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’24). Association for Computing Machine...

    2024

  3. [10]

    Faisal Kamiran and Toon Calders. 2011. Data Pre-Processing Techniques for Classification without Discrimination.Knowledge and Information Systems(2011)

    2011

  4. [11]

    Mahammed Kamruzzaman and Gene Louis Kim. 2024. Prompting Techniques for Reducing Social Bias in LLMs through System 1 and System 2 Cognitive Processes. arXiv:2404.17218 [cs.CL] https://arxiv.org/abs/2404.17218

    2024 arXiv

  5. [12]

    Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2022. Large language models are zero-shot reasoners. Advances in neural information processing systems 35 (2022), 22199–22213

    2022

  6. [13]

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Adv...

    2020

  7. [14]

    Huan Ma, Changqing Zhang, Yatao Bian, Lemao Liu, Zhirui Zhang, Peilin Zhao, Shu Zhang, Huazhu Fu, Qinghua Hu, and Bingzhe Wu. 2023. Fairness-guided Few-shot Prompting for Large Language Models. arXiv:2303.13217 [cs.CL] https: //arxiv.org/abs/2303.13217

    2023 arXiv

  8. [15]

    Evaggelia Pitoura, Panayiotis Tsaparas, Giorgos Flouris, Irini Fundulaki, Pana- giotis Papadakos, Serge Abiteboul, and Gerhard Weikum. 2018. On Measuring Bias in Online Information. SIGMOD Rec. 46, 4 (2018). https://doi.org/10.1145/ 3186549.3186553

    2018

  9. [16]

    Shaina Raza, Deepak John Reji, and Chen Ding. 2022. Dbias: Detecting biases and ensuring fairness in news articles. International Journal of Data Science and Analytics (2022), 1–21

    2022

  10. [17]

    Navid Rekabsaz, Simone Kopeinik, and Markus Schedl. 2021. Societal Biases in Retrieved Contents: Measurement Framework and Adversarial Mitigation for BERT Rankers. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieva...

    2021

  11. [18]

    Ashudeep Singh and Thorsten Joachims. 2019. Policy Learning for Fairness in Ranking. Advances in neural information processing systems 32 (2019)

    2019

  12. [20]

    Timo Spinde, Manuel Plank, Jan-David Krieger, Terry Ruas, Bela Gipp, and Akiko Aizawa. 2021. Neural Media Bias Detection Using Distant Supervision With BABE - Bias Annotations By Experts. In Findings of the Association for Computational Linguistics: EMNLP 2021. Association for...

    2021 doi

  13. [21]

    Timo Spinde, Lada Rudnitckaia, Kanishka Sinha, Felix Hamborg, Bela Gipp, and Karsten Donnay. 2021. MBIC–A Media Bias Annotation Dataset Including Annotator Characteristics. arXiv preprint arXiv:2105.11910 (2021)

    2021 arXiv

  14. [22]

    Zhenjie Xu, Wenqing Chen, Yi Tang, Xuanying Li, Cheng Hu, Zhixuan Chu, Kui Ren, Zibin Zheng, and Zhichao Lu. 2025. Mitigating Social Bias in Large Language Models: A Multi-Objective Approach within a Multi-Agent Framework. arXiv:2412.15504 [cs.CL] https://arxiv.org/abs/2412.15504

    2025

  15. [23]

    Ke Yang and Julia Stoyanovich. 2017. Measuring fairness in ranked outputs. In Proceedings of Conference on Scientific and Statistical Database Management . 1–6

    2017

  16. [24]

    Meike Zehlike, Francesco Bonchi, Carlos Castillo, Sara Hajian, Mohamed Mega- hed, and Ricardo Baeza-Yates. 2017. Fa*ir: A fair top-k ranking algorithm. In Proceedings of the 2017 ACM on Conference on Information and Knowledge Man- agement. 1569–1578

    2017

  17. [25]

    Meike Zehlike and Carlos Castillo. 2020. Reducing disparate exposure in ranking: A learning to rank approach. In Proceedings of The Web Conference . 2849–2855

    2020

  18. [26]

    Zefeng Zhang, Hengzhu Tang, Jiawei Sheng, Zhenyu Zhang, Yiming Ren, Zhenyang Li, Dawei Yin, Duohe Ma, and Tingwen Liu. 2025. Debiasing Mul- timodal Large Language Models via Noise-Aware Preference Optimization. arXiv:2503.17928 [cs.CV] https://arxiv.org/abs/2503.17928 Received...

    2025 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.