面向冻结大语言模型的学习证据强化

摘要

大型语言模型（LLMs）虽具备较强的推理能力，但在处理长文本噪声语境时常常遗漏关键证据。我们提出HiLight——一种证据强调框架，该框架将证据选择与推理过程解耦，适用于无需微调的LLM求解器。HiLight通过训练轻量级强调执行器，在保持原始语境不变的前提下为关键信息段插入最小化的高亮标记，从而避免因压缩或重写输入导致的证据丢失或扭曲。随后，冻结的求解器可基于强调后的输入进行下游推理。我们将高亮标注构建为弱监督决策问题，仅利用求解器的任务奖励通过强化学习优化执行器，无需证据标签且不修改求解器内部参数。在序列推荐和长语境问答任务上的实验表明，HiLight持续优于基于提示的基线方法和自动化提示优化方案。习得的强调策略可零样本迁移至不同规模的新求解器家族（包括基于API的求解器），表明该执行器捕捉到了真实可复用的证据结构，而非对单一骨干网络的过拟合。

English

Large Language Models (LLMs) can reason well, yet often miss decisive evidence when it is buried in long, noisy contexts. We introduce HiLight, an Evidence Emphasis framework that decouples evidence selection from reasoning for frozen LLM solvers. HiLight avoids compressing or rewriting the input, which can discard or distort evidence, by training a lightweight Emphasis Actor to insert minimal highlight tags around pivotal spans in the unaltered context. A frozen Solver then performs downstream reasoning on the emphasized input. We cast highlighting as a weakly supervised decision-making problem and optimize the Actor with reinforcement learning using only the Solver's task reward, requiring no evidence labels and no access to or modification of the Solver. Across sequential recommendation and long-context question answering, HiLight consistently improves performance over strong prompt-based and automated prompt-optimization baselines. The learned emphasis policy transfers zero-shot to both smaller and larger unseen Solver families, including an API-based Solver, suggesting that the Actor captures genuine, reusable evidence structure rather than overfitting to a single backbone.

面向冻结大语言模型的学习证据强化

Learning Evidence Highlighting for Frozen LLMs

摘要

Support