Data, Not Model: Explaining Bias toward LLM Texts in Neural Retrievers

• URL: http://arxiv.org/abs/2604.06163v1
• Date: Tue, 07 Apr 2026 17:57:07 GMT
• Title: Data, Not Model: Explaining Bias toward LLM Texts in Neural Retrievers
• Abstract summary: Recent studies show that neural retrievers often display source bias, favoring passages generated by LLMs over human-written ones.<n>This bias has been considered an inherent flaw of retrievers, raising concerns about the fairness and reliability of modern information access systems.<n>Our work challenges this view by showing that source bias stems from supervision in retrieval datasets rather than the models themselves.
• Score: 76.92519309816008
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Recent studies show that neural retrievers often display source bias, favoring passages generated by LLMs over human-written ones, even when both are semantically similar. This bias has been considered an inherent flaw of retrievers, raising concerns about the fairness and reliability of modern information access systems. Our work challenges this view by showing that source bias stems from supervision in retrieval datasets rather than the models themselves. We found that non-semantic differences, like fluency and term specificity, exist between positive and negative documents, mirroring differences between LLM and human texts. In the embedding space, the bias direction from negatives to positives aligns with the direction from human-written to LLM-generated texts. We theoretically show that retrievers inevitably absorb the artifact imbalances in the training data during contrastive learning, which leads to their preferences over LLM texts. To mitigate the effect, we propose two approaches: 1) reducing artifact differences in training data and 2) adjusting LLM text vectors by removing their projection on the bias vector. Both methods substantially reduce source bias. We hope our study alleviates some concerns regarding LLM-generated texts in information access systems.

Related papers

• Training-Induced Bias Toward LLM-Generated Content in Dense Retrieval [6.771568584669793]
  Reports claim a broad preference for text generated by large language models (LLMs)<n>In this study, we trace the emergence of such preferences across training stages and data sources.<n>Our study demonstrates that source bias is a training-induced phenomenon rather than an inherent property of dense retrievers.
  arXiv  Detail & Related papers  (2026-02-11T13:20:25Z)
• How Do LLM-Generated Texts Impact Term-Based Retrieval Models? [76.92519309816008]
  This paper investigates the influence of large language models (LLMs) on term-based retrieval models.<n>Our linguistic analysis reveals that LLM-generated texts exhibit smoother high-frequency and steeper low-frequency Zipf slopes.<n>Our study further explores whether term-based retrieval models demonstrate source bias, concluding that these models prioritize documents whose term distributions closely correspond to those of the queries.
  arXiv  Detail & Related papers  (2025-08-25T06:43:27Z)
• RepreGuard: Detecting LLM-Generated Text by Revealing Hidden Representation Patterns [50.401907401444404]
  Large language models (LLMs) are crucial for preventing misuse and building trustworthy AI systems.<n>We propose RepreGuard, an efficient statistics-based detection method.<n> Experimental results show that RepreGuard outperforms all baselines with average 94.92% AUROC on both in-distribution (ID) and OOD scenarios.
  arXiv  Detail & Related papers  (2025-08-18T17:59:15Z)
• Reverse Modeling in Large Language Models [39.18082267850437]
  Humans are accustomed to reading and writing in a forward manner.<n>This paper investigates whether auto-regressive large language models (LLMs) struggle with reverse modeling.
  arXiv  Detail & Related papers  (2024-10-13T12:24:03Z)
• ReMoDetect: Reward Models Recognize Aligned LLM's Generations [55.06804460642062]
  Large language models (LLMs) generate human-preferable texts. In this paper, we identify the common characteristics shared by these models. We propose two training schemes to further improve the detection ability of the reward model.
  arXiv  Detail & Related papers  (2024-05-27T17:38:33Z)
• White Men Lead, Black Women Help? Benchmarking and Mitigating Language Agency Social Biases in LLMs [58.27353205269664]
  Social biases can manifest in language agency in Large Language Model (LLM)-generated content.<n>We introduce the Language Agency Bias Evaluation benchmark, which comprehensively evaluates biases in LLMs.<n>Using LABE, we unveil language agency social biases in 3 recent LLMs: ChatGPT, Llama3, and Mistral.
  arXiv  Detail & Related papers  (2024-04-16T12:27:54Z)
• Reinforcement Learning from Multi-role Debates as Feedback for Bias Mitigation in LLMs [6.090496490133132]
  We propose Reinforcement Learning from Multi-role Debates as Feedback (RLDF), a novel approach for bias mitigation replacing human feedback in traditional RLHF. We utilize LLMs in multi-role debates to create a dataset that includes both high-bias and low-bias instances for training the reward model in reinforcement learning.
  arXiv  Detail & Related papers  (2024-04-15T22:18:50Z)
• Pride and Prejudice: LLM Amplifies Self-Bias in Self-Refinement [75.7148545929689]
  Large language models (LLMs) improve their performance through self-feedback on certain tasks while degrade on others. We formally define LLM's self-bias - the tendency to favor its own generation. We analyze six LLMs on translation, constrained text generation, and mathematical reasoning tasks.
  arXiv  Detail & Related papers  (2024-02-18T03:10:39Z)
• Do LLMs exhibit human-like response biases? A case study in survey design [66.1850490474361]
  We investigate the extent to which large language models (LLMs) reflect human response biases, if at all. We design a dataset and framework to evaluate whether LLMs exhibit human-like response biases in survey questionnaires. Our comprehensive evaluation of nine models shows that popular open and commercial LLMs generally fail to reflect human-like behavior.
  arXiv  Detail & Related papers  (2023-11-07T15:40:43Z)
• Neural Retrievers are Biased Towards LLM-Generated Content [35.40318940303482]
  Large language models (LLMs) have revolutionized the paradigm of information retrieval (IR) applications. How these LLM-generated documents influence the IR systems is a pressing and still unexplored question. Surprisingly, our findings indicate that neural retrieval models tend to rank LLM-generated documents higher.
  arXiv  Detail & Related papers  (2023-10-31T14:42:23Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.