arXiv cs.MA· Sonali, Deeksha Varshney, Asif Ekbal·· 11 小时前AI 评分26
ReSUM：面向查询聚焦摘要的推理驱动多智能体框架
ReSUM: Reasoning-Driven Multi-Agent Framework for Query-Focused Summarization
AI 导读
ReSUM 是一个由 planner 协调检索与摘要智能体的模块化多 LLM 框架，用于解决长会议查询聚焦摘要（QFS）中的信息遗漏与幻觉问题。该框架将教师推理蒸馏进以查询和计划为条件的检索器与摘要器，提升长上下文鲁棒性并减少无关探索。在查询聚焦会议摘要基准上取得一致提升。
正文 · 原文
Abstract:LLMs excel at language tasks, but long-meeting QFS remains difficult due to complex speaker dynamics, sparse key information, and limited data. Meetings involve disfluencies, long transcripts and complex discourse. QFS have high risk of information omission, hallucination, even for advanced models such as GPT-4-mini and Gemini. Existing end-to-end methods compress retrieval, evidence extraction, and summary generation into a single pass, failing to handle the complex reasoning required for targeted meeting summaries. We propose ReSUM, a modular multi-LLM framework coordinated by a planner that orchestrates retrieval and summarization agents. By distilling teacher reasoning into query- and plan-conditioned retrievers and summarizers, ReSUM enhances long-context robustness, minimizes irrelevant exploration, and generates more faithful and comprehensive query-focused summaries. Experiments show consistent gains on query-focused meeting summarization benchmarks.
| Comments: | 10 pages, 4 figures |
| Subjects: | Multiagent Systems (cs.MA) |
| Cite as: | arXiv:2611.09349 [cs.MA] |
| (or arXiv:2611.09349v1 [cs.MA] for this version) | |
| https://doi.org/10.48550/arXiv.2611.09349 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Sonali Sonali [view email]
[v1]
Fri, 6 Nov 2026 11:45:14 UTC (476 KB)
来源：arXiv cs.MA · arxiv.org