Reproduce “TraceMem: Weaving Narrative Memory Schemata from User Conversational Traces” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yiming Shu, Pei Liu, Tiange Zhang, Ruiyang Gao, Jun Ma, Chen Sun
TraceMem structures user conversational traces into evolving narrative memory schemata and user memory cards for large language models. The framework, using a three-stage memory consolidation pipeline and agentic retrieval, achieved up to 17% higher overall accuracy than FullText baselines on the LoCoMo benchmark, particularly excelling in multi-hop and temporal reasoning tasks.
Key insights
- Organizing conversational histories into coherent, time-evolving narrative threads and structured user memory cards significantly enhances LLM long-term conversational intelligence and persona consistency.
- A cognitively-inspired memory consolidation process, transforming fragmented interactions into personal semantics, improves reasoning capabilities, especially for complex multi-hop and temporal tasks.
- The Agentic Search mechanism is crucial for efficiently navigating and synthesizing complex histories by judiciously retrieving relevant narrative threads, mimicking human recall.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.