SpeakerMem-R1: Speaker-Centered Dual-Track Memory for Multi-Party Dialogue
SpeakerMem-R1 stores speaker-labeled messages and structured states, reaching 62.33% on EverMemBench.
SpeakerMem-R1 targets multi-party conversational memory with speaker-labeled verbatim messages and derived states organized into person-level and group-level views, combined by entity, event, and time. Writer-R1 is trained with SpeakerLevenshtein and speaker-conditioned GRPO to reduce attribution and update errors during memory construction. Reported binary accuracies are 47.9% on GroupMemBench, 69.2% on SocialMemBench, and 61.9% on EverMemBench, with 62.33% on the public EverMind-AI EverMemBench leaderboard and 70.85% on 1,986 LoCoMo questions. On 305 controlled questions, reinforcement learning raises the supervised Writer's mean accuracy from 57.38% to 68.20%.
- Dual-track memory stores verbatim speech and derived states
- Person and group views are combined by entity, event, and time
- Public EverMemBench score of 62.33% leads reported results
- RL raises Writer accuracy from 57.38% to 68.20%
- Organizations
- EverMind-AI
- AI models
- SpeakerMem-R1Writer-R1
From the original article
1 min read at huggingface.co
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.
By Haobo Zheng, Tan Tang, Yan Chen · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.