SpeakerMem-R1: speaker-centered dual-track memory for multi-party dialogue
Zhejiang University
Builds long-term memory for multi-party conversations around who-said-what and relational state: one track stores verbatim speaker-labeled messages, the other person-level and group-level derived states. The Writer module is RL-trained (GRPO) with a SpeakerLevenshtein reward to cut attribution errors, claiming SOTA on EverMemBench (62.33%).
Why it matters
Tops the non-embodied half of the Sep 24 HF Daily board (~70-79 upvotes), targeting a concrete unsolved failure mode: attribution and social-state tracking in group chat memory.
Importance: 2/5
Notable HF Daily paper (~70-79 upvotes) with SOTA claim
Sources
official
HuggingFace Daily Papers entry