Research map / Large language models / Retrieval and knowledge
HGMem: Hypergraph-based Working Memory to Improve Multi-step RAG for Long-Context Complex Relational Modeling
Chulun Zhou, Chunkang Zhang, Guoxin Yu, Fandong Meng, Jie Zhou, Wai Lam, Mo Yu
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 merging operation to memory evolution (ablation) | HGMem (Qwen2.5-32B) · NarrativeQA | accuracy: 43.00 -> 51.00 | Improved |
| add merging operation to memory evolution (ablation) | HGMem (Qwen2.5-32B) · Longbench | diversity: 61.80 -> 66.51 | Improved |
| add merging operation to memory evolution (ablation) | HGMem (Qwen2.5-32B) · Prelude | accuracy: 57.78 -> 62.22 | Improved |
| add update operation to memory evolution (ablation) | HGMem (Qwen2.5-32B) · Longbench | comprehensiveness: 62.48 -> 64.18 | Improved |
| add update operation to memory evolution (ablation) | HGMem (Qwen2.5-32B) · NarrativeQA | accuracy: 50.00 -> 51.00 | Improved |
| use adaptive retrieval instead of Global Exploration only (ablation) | HGMem (Qwen2.5-32B) · Longbench | comprehensiveness: 59.25 -> 64.18 | Improved |
| use adaptive retrieval instead of Local Investigation only (ablation) | HGMem (Qwen2.5-32B) · NoCha | accuracy: 63.49 -> 70.63 | Improved |
| add merging on sense-making queries (ablation) | HGMem (Qwen2.5-32B) · NarrativeQA sense-making queries | accuracy: 30.00 -> 40.00 | Improved |
| add merging on sense-making queries (ablation) | HGMem (Qwen2.5-32B) · NarrativeQA sense-making queries | avg entities per hyperedge: 4.10 -> 7.07 | Improved |
| add merging on primitive queries (ablation) | HGMem (Qwen2.5-32B) · NoCha primitive queries | accuracy: 65.00 -> 60.00 | Hurt |
| replace offline graph with 50% ablated graph | HGMem (Qwen2.5-32B) · Prelude | accuracy: 62.22 -> 57.97 | Hurt |
| replace LLM-built graph with Stanford OpenIE graph | HGMem (Qwen2.5-32B) · NoCha | accuracy: 70.63 -> 66.67 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeepRAG — DeepRAG: Thinking to Retrieve Step by Step for Large Language Models (ICLR 2026) · better on 10 of 10 shared results
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
- HyperGraphRAG: Retrieval-Augmented Generation via Hypergraph-Structured Knowledge Representation (NeurIPS 2025 · similarity 0.88)
- Simple is Effective: The Roles of Graphs and Large Language Models in Knowledge-Graph-Based Retrieval-Augmented Generation (ICLR 2025 · similarity 0.88)
- Memory is Reconstructed, Not Retrieved: Graph Memory for LLM Agents (ICML 2026 · similarity 0.87)
- StructRAG: Boosting Knowledge Intensive Reasoning of LLMs via Inference-time Hybrid Information Structurization (ICLR 2025 · similarity 0.87)
- From Isolated Conversations to Hierarchical Schemas: Dynamic Tree Memory Representation for LLMs (ICLR 2025 · similarity 0.87)
- Think-on-Graph 2.0: Deep and Faithful Large Language Model Reasoning with Knowledge-guided Retrieval Augmented Generation (ICLR 2025 · 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.
