AutoViewMem is a data‑driven framework that first organizes conversational long‑term memory into self‑configuring, low‑overlap semantic views during the write phase, then indexes them. The system discovers candidate views from interaction traces, selects a compact complementary set, and uses these views to guide provenance‑grounded structured extraction at write time, producing traceable memory entries.
By shifting semantic disentanglement from retrieval to write time, a standard top‑K similarity search can retrieve focused evidence without explicit routing or iterative retrieval. An offline consolidation step further improves memory compactness and consistency.
Experiments on the LoCoMo and PersonaMem benchmarks with Qwen3‑8B and Qwen3‑14B backbones show that AutoViewMem outperforms strong memory baselines on long‑horizon QA and personalization while preserving a simple inference pipeline. Review