[论文] [论文] SpeakerMem-R1: Speaker-Centered Dual-Track Memory for Multi-...

论文概要

研究领域: NLP 作者: Haobo Zheng, Tan Tang, Yan Chen, Weijie Wang, Yingcai Wu 发布时间: 2026-09-22 arXiv: 2609.26780

中文摘要

多方场景下的长期对话记忆，不止是从长期对话中检索相关内容：须区分谁说了什么、每句涉及谁、彼此如何看待、群体共享什么、状态如何随时间变化。近期基准研究显示，现有通用 LLM 记忆系统常丢失人物与群体关系，或难以整合分布在成员、群体与时间上的线索——两大核心瓶颈：多方对话中的消息归属与关系理解，以及从交错历史重建状态。为此提出 SpeakerMem-R1：双轨记忆存储带说话人标签的原始消息与派生状态，组织为个人级与群体级视图，查询时按实体、事件、时间合并两轨证据。为减少归属/更新错误并支持本地部署，用 SpeakerLevenshtein 与说话人条件化 GRPO 训练 Writer-R1。在 GroupMemBench、SocialMemBench、EverMemBench 上二分类准确率分别为 47.9%、69.2%、61.9%；在 EverMind-AI 公布的 EverMemBench 榜单取得 62.33%，为现有最佳报告结果；全部 1,986 个 LoCoMo 问题（两人长期对话边界测试）上 70.85%。305 题控制评测中，RL 将 SFT Writer 平均准确率从 57.38% 提至 68.20%。消融显示原始文本轨与结构化轨、个人级与群体级视图互为补充。

原文摘要

Long-term conversational memory in multi-party settings requires more than retrieving relevant content from long-term conversations: it must distinguish who said what, whom each statement concerns, how individuals perceive one another, what information is shared by the group, and how states change over time. Recent studies on multi-party dialogue benchmarks show that existing general-purpose LLM memory systems tend to lose person and group relations or struggle to integrate clues distributed across members, groups, and time. Together, these issues reveal two core bottlenecks: message attribution and relational understanding in multi-party dialogue, and state reconstruction from interleaved histories. To address both, we propose \(\textbf{SpeakerMem-R1}\): its dual-track memory stores speaker-labeled verbatim messages and derived states organized into person-level and group-level views, then combines evidence from both tracks by entity, event, and time at query time. To reduce attribution and update errors during structured memory construction while enabling local deployment, we train Writer-R1 with SpeakerLevenshtein and speaker-conditioned GRPO. On GroupMemBench, SocialMemBench, and EverMemBench, SpeakerMem-R1 achieves binary accuracies of 47.9%, 69.2%, and 61.9%, respectively. On the publicly reported EverMemBench leaderboard from EverMind-AI, we achieves 62.33%, the best reported result among the latest state-of-the-art frameworks. It also achieves 70.85% on all 1,986 LoCoMo questions, which we use as a two-person long-term conversation boundary test. In a controlled evaluation of 305 questions, RL raises the SFT Writer's mean accuracy from 57.38% to 68.20%. We report both binary accuracy and token-F1, and ablations show that the verbatim and structured tracks, as well as person-level and group-level views, are complementary under the standardized evaluation interface.

---

*自动采集于 2026-09-24*

#论文 #arXiv #NLP #小凯