Research map / Graph learning / Graph self-supervised learning
Efficient Code Analysis via Graph Representation Learning-Guided Large Language Models
Hang Gao, Tao Peng, Baoquan Cui, Hong Huang, Fengge Wu, Junsuo Zhao, Jian Zhang
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 GNN-attention subgraph guidance to the LLM | GPT-4o · MalCP (All) | accuracy (%): 90.39 -> 95.62 | Improved |
| add GNN-attention subgraph guidance to the LLM | GPT-4o · MalCP (Large) | accuracy (%): 77.98 -> 90.41 | Improved |
| add GNN-attention subgraph guidance to the LLM | Llama 3 · MalCP (All) | accuracy (%): 35.60 -> 81.60 | Improved |
| add GNN-attention subgraph guidance to the LLM | GPT-4o · MalCP (Large) | mean tokens: 251001.56 -> 640.06 | Improved |
| add GNN-attention guidance on small packages | GPT-4o · MalCP (Small) | recall (%): 97.38 -> 99.34 | No clear effect |
| add GMLLM guidance vs best specialized detector Ea4mp | GPT-4o · Datadog | accuracy (%): 94.67 -> 96.96 | Mixed |
| add common human rules Scomm to structure-only graph (ablation) | GMLLM (GPT-4o) · MalCP (All) | accuracy (%): 88.65 -> 91.86 | Improved |
| add data-derived rules Sdata to Scomm (ablation) | GMLLM (GPT-4o) · MalCP (All) | accuracy (%): 91.86 -> 95.62 | Improved |
| add LLM analysis on top of GNN-only classifier (ablation) | GMLLM (GPT-4o) · MalCP (All) | accuracy (%): 92.22 -> 95.62 | Improved |
| increase edge budget K from 10 to 20 (ablation) | GMLLM (GPT-4o) · MalCP | accuracy (%): 92.97 -> 95.62 | Improved |
| increase edge budget K beyond 20 (ablation) | GMLLM (GPT-4o) · MalCP | accuracy (%): 95.62 -> 94.88 | No clear effect |
| increase edge budget K from 20 to 50 (ablation) | GMLLM (GPT-4o) · MalCP (Large) | token cost: 640.06 -> 1131.33 | Hurt |
Closest papers by meaning
- Code Graph Model (CGM): A Graph-Integrated Large Language Model for Repository-Level Software Engineering Tasks (NeurIPS 2025 · similarity 0.86)
- SecCodePRM: A Process Reward Model for Code Security (ICML 2026 · similarity 0.85)
- CodeMimicry: Exploiting Safety Generalization Lag in Large Language Models via Structured Code Completion (NeurIPS 2026 · similarity 0.85)
- CODE REPRESENTATION LEARNING AT SCALE (ICLR 2024 · similarity 0.84)
- Evaluation of Large Language Models on Code Obfuscation (Student Abstract) (AAAI 2024 · similarity 0.84)
- Local Sparsity Enables Unsupervised LLM Safety Detection (NeurIPS 2026 · similarity 0.84)
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.
