发表机构
University of Electronic Science and Technology of China; Tsinghua University; Institute of Information Engineering, Chinese Academy of Sciences; University of Chinese Academy of Sciences; Xiaomi Inc.; MiLM Plus（电子科技大学; 清华大学; 中国科学院信息工程研究所; 中国科学院大学; 小米公司; MiLM Plus）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
提出MemLoc框架，通过检索-定位-生成流程解决长期对话记忆问答中的证据分散和噪声问题，实现最先进的检索准确性和响应质量。
AI 中文摘要
检索增强生成（RAG）使大型语言模型（LLMs）能够通过访问外部知识来回答问题，并已被广泛用于长期对话记忆问答。然而，现有方法面临两个关键挑战：（1）证据碎片分散在时间上相隔较远的会话中，（2）检索到的会话中的噪声内容触发了“中间丢失”效应。为了解决这些挑战，我们提出了MemLoc，一个统一的检索-定位-生成框架，用于长期对话记忆问答。对于检索，MemLoc将每个会话分解为多粒度记忆单元，并通过基于熵的粒度选择的内记忆图进行查询路由。它进一步通过跨记忆图建模跨会话的语义和时间依赖，实现从粗到细的检索，获取前K个相关记忆候选。对于定位，我们引入了一个基于推理的证据定位器，使用自我反思提示策略优化（SHPO）进行训练，通过提取记忆单元中与查询相关的片段来逐步细化以抑制噪声，并在候选之间重新排序以去除冗余，生成带有轻量级位置ID的紧凑证据集。对于生成，这些ID作为精确的接地信号，引导LLM到正确的记忆位置，减轻“中间丢失”效应，同时保持原始上下文的完整性。在四个基准上的大量实验表明，MemLoc在保持效率的同时实现了最先进的检索准确性和响应质量。我们的代码可在以下网址获取：this https URL。
英文摘要
Retrieval-augmented generation (RAG) enables large language models (LLMs) to answer questions by accessing external knowledge and has been widely adopted for long-term conversational memory question answering. However, existing methods suffer from two key challenges: (1) fragmented evidence scattered across temporally distant sessions, and (2) noisy content within retrieved sessions that triggers the lost-in-the-middle effect. To address these challenges, we propose MemLoc, a unified Retrieve-Localize-Generate framework for long-term conversational memory QA. For retrieval, MemLoc decomposes each session into multi-granularity memory units and performs query routing via an inner-memory graph with entropy-based granularity selection. It further models cross-session semantic and temporal dependencies through a cross-memory graph, enabling coarse-to-fine retrieval of top-K relevant memory candidates. For localization, we introduce a reasoning-based evidence locator trained with Self-reflective Hint Policy Optimization (SHPO), which performs progressive refinement by extracting query-relevant fragments within memory units to suppress noise and reranking across candidates to remove redundancy, producing a compact evidence set with lightweight location IDs. For generation, these IDs act as precise grounding signals that guide the LLM to the correct memory positions, mitigating the lost-in-the-middle effect while preserving original contextual integrity. Extensive experiments on four benchmarks demonstrate that MemLoc achieves state-of-the-art retrieval accuracy and response quality while maintaining efficiency. Our code is available at: https://github.com/Nikol-coder/MemLoc.
Comments22 pages, 4 figures, 14 tables. Accepted to the EMNLP 2026 Main Conference
