AutoViewMem is a framework that organizes long-term conversational memory into self-configuring, low-overlap semantic views before indexing. It discovers candidate views from interaction traces, selects a compact complementary set, and uses them to guide write-time structured extraction of provenance-grounded memories.
Experiments on the LoCoMo and PersonaMem benchmarks with Qwen3-8B and Qwen3-14B backbones show that AutoViewMem improves long-horizon question answering and personalization over existing memory systems like Mem0, MemGAS, MemoryBank, and A-mem. The method preserves a simple inference pipeline without explicit routing or iterative retrieval.
By moving semantic disentanglement from retrieval time to write time, AutoViewMem reduces semantic interference in mixed-content memories, allowing standard top-K similarity search to retrieve focused evidence. This representation-first design addresses a core limitation of current memory systems that often struggle with heterogeneous information embedded in single representations.