Research map / Large language models / Retrieval and knowledge
From Single to Multi-Granularity: Toward Long-Term Memory Association and Selection of Conversational Agents
Derong Xu, Yi Wen, Pengyue Jia, Yingyi Zhang, Wenlin Zhang, Yichao Wang, Huifeng Guo, Ruiming Tang, Xiangyu Zhao, Enhong Chen, Tong Xu
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 |
|---|---|---|---|
| add Gaussian-mixture memory association (ablation) | MemGAS · LongMemEval-s | GPT4o-J: 57.20 -> 60.20 | Improved |
| add Personalized PageRank retrieval over the association graph (ablation) | MemGAS · LongMemEval-s | GPT4o-J: 56.60 -> 60.20 | Improved |
| add entropy-based granularity router (ablation) | MemGAS · LongMemEval-s | F1: 18.88 -> 20.38 | Improved |
| add memory association (GMM + PPR) (ablation) | MemGAS · LongMemEval-s | Recall@3: 74.89 -> 78.51 | Improved |
| enable all four modules together vs none (ablation) | MemGAS · LongMemEval-s | F1: 13.78 -> 20.38 | Improved |
| use router and association instead of naive combination of granularities (ablation) | multi-granularity retrieval · LongMemEval-s | F1: 14.59 -> 20.38 | Improved |
| use multiple granularities (naive combination) instead of session-level only (ablation) | Contriever retrieval · LongMemEval-s | F1: 13.78 -> 14.59 | Improved |
| retrieve with summary-level or keyword-level units only (ablation) | Contriever retrieval · LongMemEval-s | GPT4o-J: 55.40 -> 28.80 | Hurt |
| retrieve with turn-level units instead of session-level (ablation) | Contriever retrieval · LongMemEval-s | GPT4o-J: 55.40 -> 57.60 | Improved |
| increase retrieved Top-K to high values (ablation) | MemGAS · LongMemEval-s | F1 | Hurt |
| use MemGAS instead of full history context | GPT-4o-mini QA · LongMemEval-s | GPT4o-J: 50.60 -> 60.20 | Improved |
| MemGAS vs HippoRAG 2 on LoCoMo GPT4o-J | MemGAS · LoCoMo | GPT4o-J: 45.62 -> 41.07 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- HippoRAG 2 — HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models (NeurIPS 2024) · better on 4 of 5 shared results
- RAPTOR — RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval (ICLR 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SeCom: On Memory Construction and Retrieval for Personalized Conversational Agents (ICLR 2025 · similarity 0.90)
- LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (ICLR 2025 · similarity 0.90)
- Evaluating Memory in LLM Agents via Incremental Multi-Turn Interactions (ICLR 2026 · similarity 0.88)
- From Isolated Conversations to Hierarchical Schemas: Dynamic Tree Memory Representation for LLMs (ICLR 2025 · similarity 0.88)
- MemGuide: Intent-Driven Memory Selection for Goal-Oriented Multi-Session LLM Agents (AAAI 2026 · similarity 0.88)
- PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents (ICML 2026 · similarity 0.87)
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.
