Research map / Large language models / Retrieval and knowledge
Evoking User Memory: Personalizing LLM via Recollection-Familiarity Adaptive Retrieval
Yingyi Zhang, Junyi Li, Wenlin Zhang, Pengyue Jia, Xianneng Li, Yichao Wang, Derong Xu, Yi Wen, Huifeng Guo, Yong Liu, Xiangyu Zhao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| adaptive switching between familiarity and recollection instead of dense top-K only (ablation) | RF-Mem · PersonaMem 32K | overall accuracy: 0.5908 -> 0.6350 | Improved |
| adaptive switching instead of dense top-K only at 1M corpus (ablation) | RF-Mem · PersonaMem 1M | overall accuracy: 0.4518 -> 0.4589 | Improved |
| add entropy-guided gating over always-on recollection (ablation) | RF-Mem · PersonaMem 32K | overall accuracy: 0.6214 -> 0.6350 | Improved |
| add entropy-guided gating over always-on recollection (latency) (ablation) | RF-Mem · PersonaMem 32K | retrieval time: 7.09ms -> 5.09ms | Improved |
| retrieve with RF-Mem instead of feeding full context | RF-Mem · PersonaMem 128K | overall accuracy: 0.3231 -> 0.5394 | Improved |
| retrieve with RF-Mem instead of feeding full context at 32K | RF-Mem · PersonaMem 32K | overall accuracy: 0.6129 -> 0.6350 | Improved |
| RF-Mem vs familiarity on PersonaBench | RF-Mem · PersonaBench (MiniLM) | overall Recall@10: 0.5964 -> 0.6071 | Improved |
| RF-Mem vs familiarity on LongMemEval-M | RF-Mem · LongMemEval-M (MiniLM) | Recall@5: 0.4177 -> 0.4391 | Improved |
| layer RF-Mem on MemoryBank summary index | RF-Mem · PersonaMem 32K (MemoryBank summary index) | overall accuracy: 0.4941 -> 0.5314 | Improved |
| use RF-Mem as retrieval layer in Search-o1 | Search-o1 · PersonaMem 32K (Search-o1) | overall accuracy: 0.5823 -> 0.6146 | Improved |
| RF-Mem on HyDE-expanded queries | HyDE retrieval · PersonaBench (HyDE expansion, MiniLM) | overall Recall@10: 0.5120 -> 0.5194 | No clear effect |
| increase mixing alpha (more weight on original query) (ablation) | Recollection retrieval · LongMemEval-M | Recall@5 / Recall@50 | Mixed |
Closest papers by meaning
- Human-inspired Episodic Memory for Infinite Context LLMs (ICLR 2025 · similarity 0.89)
- Personalize-then-Store: Benchmarking and Learning Personalized Memory for Long-Horizon Agents (NeurIPS 2026 · similarity 0.88)
- From RAG to Memory: Non-Parametric Continual Learning for Large Language Models (ICML 2025 · similarity 0.87)
- REMem: Reasoning with Episodic Memory in Language Agent (ICLR 2026 · similarity 0.87)
- HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models (NeurIPS 2024 · similarity 0.87)
- Memory Type Varies: Empowering LLM Agents for Long-Term Memory with Diverse Strategies (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
