[论文] Context-Driven Incremental Compression for Multi-Turn Dialogue Ge...

论文概要

研究领域: NLP
作者: Yeongseo Jung, Jaehyeok Kim, Eunseo Jung, Jiachuan Wang, Yongqi Zhang, Ka Chun Cheung, Simon See, Lei Chen
发布时间: 2026-06-10
arXiv: 2606.12411

中文摘要

现代对话智能体在每一轮都以不断增长的对话历史为条件，导致冗余的注意力与编码成本随对话长度增长。朴素截断或摘要会降低保真度，而现有上下文压缩器缺乏跨轮记忆共享或修正，导致信息损失和长期对话中的复合错误。我们重新审视对话动态下的上下文压缩，并实证呈现其脆弱性。为提高效率和鲁棒性，我们引入上下文驱动增量压缩（C-DIC），将对话视为交织的上下文线程，将可修正的每线程压缩状态存储在单一紧凑对话记忆中。每轮中，一个轻量级的检索-修正-写回循环跨轮共享信息并更新陈旧记忆，稳定长期行为。此外，我们将截断时间反向传播（TBPTT）适应到多轮设置，无需完整历史反向传播即可学习跨轮依赖。在长篇对话基准上的大量实验证明了C-DIC的优越性能和效率；值得注意的是，C-DIC在数百轮对话中显示稳定的推理延迟和困惑度，支持高质量对话建模的可扩展路径。

原文摘要

Modern conversational agents condition on an ever-growing dialogue history at each turn, incurring redundant attention and encoding costs that grow with conversation length. Naive truncation or summarization degrades fidelity, while existing context compressors lack cross-turn memory sharing or revision, causing information loss and compounding errors in long dialogues. We revisit the context compression under conversational dynamics and empirically present its fragility. To improve both efficiency and robustness, we introduce Context-Driven Incremental Compression (C-DIC), which treats a conversation as interleaved contextual threads and stores revisable per-thread compression states in a single, compact dialogue memory. At each turn, a lightweight retrieve, revise, and write-back loop sh...

自动采集于 2026-06-12

#论文 #arXiv #NLP #小凯