发现检索模型偏爱大模型文本，根源在训练数据而非模型本身
Data, Not Model: Explaining Bias toward LLM Texts in Neural Retrievers
- 指出检索偏差源于训练数据中非语义特征差异
- 实验证明大模型文本与人工文本在流畅性等特征上存在系统性差异
- 提出数据去偏和向量修正两种方法，显著降低偏见
近期研究发现，神经检索器常表现出源偏见，倾向于选择大语言模型生成的段落而非人工撰写的段落，即使两者语义相似。这种偏见曾被视为检索器固有缺陷，引发对现代信息获取系统公平性和可靠性的担忧。本文挑战这一观点，指出源偏见源于检索数据集中的监督信号，而非模型本身。我们发现正负样本间存在非语义差异，如流畅性和术语特异性，这些差异与大模型与人工文本的差异一致。在嵌入空间中，从负样本到正样本的偏见方向，与从人工文本到大模型文本的方向一致。理论上证明，在对比学习过程中，检索器不可避免地吸收训练数据中的特征失衡，从而偏好大模型文本。为此，我们提出两种缓解方法：1）减少训练数据中的特征差异；2）通过移除大模型文本向量在偏见方向上的投影进行修正。两种方法均显著降低源偏见。希望本研究能缓解人们对大模型生成文本在信息检索中应用的担忧。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。