返回全部动态
密集检索模型性别敏感性的机制分析
原标题：A Mechanistic Analysis of Gender Sensitivity in Dense Retrieval Models
AI 摘要
该研究对密集检索模型中的性别敏感性进行了机制性分析，发现性别信号源于输入嵌入，并通过少数晚期注意力头传播，这些注意力头同时携带性别和术语匹配信号。基于此，研究者测试了嵌入级和注意力级干预，发现嵌入级干预非特异性地中和分数差异，而注意力级干预产生方向性偏移。研究为针对性去偏提供了机制基础，并强调了在共享模型组件中分离性别与相关性信号的挑战。
以上摘要由 AI 生成，可能存在误差。事实请以原文为准。
正文节选
Computer Science > Information Retrieval Title:A Mechanistic Analysis of Gender Sensitivity in Dense Retrieval Models View PDF HTML (experimental) Abstract:While gender bias in dense retrieval models is well documented, with prior work showing that models often score male-gendered documents higher than female or neutral variants, the internal mechanisms producing these disparities are poorly understood. In this paper, we mechanistically analyze bi-encoder models to localize gender se
发布时间：2026-08-07 12:00
抓取时间：2026-08-07 15:29
来源机构：arXiv