arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.12727cs.IR
收藏

利用多正例段落训练密集检索器

Training Dense Retrievers with Multiple Positive Passages

University of Chinese Academy of Sciences(中国科学院大学)

Benben Wang, Minghao Tang, Hengran Zhang, Jiafeng Guo, Keping Bi

AI总结:

本文提出利用多正例段落训练检索器,通过统一对比学习框架分析不同优化目标,发现LSEPair在鲁棒性和性能上表现优异,同时揭示了不同目标对正样本质量的敏感性。

摘要:

现代知识密集型系统,如检索增强生成(RAG),依赖于有效的检索器来为下游模块设定性能上限。然而,检索器训练受到稀疏、单正例标注的瓶颈,导致假负样本噪声和次优监督。尽管大语言模型(LLMs)的出现使收集大规模全面的多正例相关标签成为可能,但如何将这些密集信号最优地融入训练仍缺乏深入理解。在本文中,我们对检索器训练中的多正例优化目标进行了系统研究。我们统一了代表性的目标,包括联合似然(JointLH)、求和边际似然(SumMargLH)和对数和指数对称(LSEPair)损失,纳入共享对比学习框架。我们的理论分析刻画了它们不同的梯度行为,揭示了每个目标如何在正文档集上分配概率质量。实证上,我们在自然问题、MS MARCO和BEIR基准上进行了广泛评估,涵盖两种现实场景:同质LLM标注数据和混合的人类和LLM标签。我们的结果表明,LSEPair在各种设置中始终表现出优越的鲁棒性和性能,而JointLH和SumMargLH对正样本质量高度敏感。此外,我们发现简单的随机采样(Rand1LH)策略是一个可靠的基线。通过将理论洞察与实证发现对齐,我们为利用密集的LLM增强监督来增强检索器效果提供了实用的设计原则。

英文摘要:

Modern knowledge-intensive systems, such as retrieval-augmented generation (RAG), rely on effective retrievers to establish the performance ceiling for downstream modules. However, retriever training has been bottlenecked by sparse, single-positive annotations, which lead to false-negative noise and suboptimal supervision. While the advent of large language models (LLMs) makes it feasible to collect comprehensive multi-positive relevance labels at scale, the optimal strategy for incorporating these dense signals into training remains poorly understood. In this paper, we present a systematic study of multi-positive optimization objectives for retriever training. We unify representative objectives, including Joint Likelihood (JointLH), Summed Marginal Likelihood (SumMargLH), and Log-Sum-Exp Pairwise (LSEPair) loss, under a shared contrastive learning framework. Our theoretical analysis characterizes their distinct gradient behaviors, revealing how each allocates probability mass across positive document sets. Empirically, we conduct extensive evaluations on Natural Questions, MS MARCO, and the BEIR benchmark across two realistic regimes: homogeneous LLM-annotated data and heterogeneous mixtures of human and LLM labels. Our results show that LSEPair consistently achieves superior robustness and performance across settings, while JointLH and SumMargLH exhibit high sensitivity to the quality of positives. Furthermore, we find that the simple strategy of random sampling (Rand1LH) serves as a reliable baseline. By aligning theoretical insights with empirical findings, we provide practical design principles for leveraging dense, LLM-augmented supervision to enhance retriever effectiveness.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑