AutoViewMem: Self-Configuring Orthogonal Views for Conversational Long-Term Memory
AutoViewMem shifts memory cleanup to write time, splitting mixed conversation history into low-overlap semantic views before retrieval.
The paper says fixed schemas and single mixed memories can bury relevant evidence when preferences, events, constraints, and updates collide. Its framework discovers candidate views from interaction traces, selects a compact complementary set, and uses them for structured, provenance-grounded memory extraction. The authors report gains on LoCoMo and PersonaMem with Qwen3-8B and Qwen3-14B, while keeping inference to standard top-K similarity search. ArXiv · AI/CL/LG's note
The paper says fixed schemas and single mixed memories can bury relevant evidence when preferences, events, constraints, and updates collide. Its framework discovers candidate views from interaction traces, selects a compact complementary set, and uses them for structured, provenance-grounded memory extraction. The authors report gains on LoCoMo and PersonaMem with Qwen3-8B and Qwen3-14B, while keeping inference to standard top-K similarity search. ArXiv · AI/CL/LG's note
score 5