Research map / Graph learning / Graph self-supervised learning
Graph-Oriented Instruction Tuning of Large Language Models for Generic Graph Mining
Yanchao Tan, Hang Lv, Pengxiang Zhan, Shiping Wang, Carl Yang
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 CoT-based instructions to NC instruction package (ablation) | MuseGraph (LLaMA3-8B) · IMDB | Micro-F1: 71.23 -> 73.03 | Improved |
| add CoT-based instructions to NC instruction package (ablation) | MuseGraph (LLaMA3-8B) · IMDB | Macro-F1: 70.62 -> 71.45 | Improved |
| raise LP CoT package allocation from 1:1 to 3:1 with NC (ablation) | MuseGraph (LLaMA3-8B) · IMDB | Micro-F1: 72.81 -> 74.86 | Improved |
| diversify instruction packages across multiple tasks and datasets (full MuseGraph) (ablation) | MuseGraph (LLaMA3-8B) · IMDB | Micro-F1: 74.86 -> 76.57 | Improved |
| integrate all one-hop neighbors versus central node attributes only (ablation) | MuseGraph · IMDB | Macro-F1: up to 13.34% | Improved |
| extend neighborhood to two-hop or three-hop without sampling (ablation) | MuseGraph · IMDB | Micro-F1 | Hurt |
| one-hop neighbors plus random walks with node-energy selection versus plain multi-hop neighbors (ablation) | MuseGraph · IMDB | Micro-F1 | Improved |
| use LLaMA3-8B instead of other foundation models (ablation) | MuseGraph · multiple tasks | accuracy and BLEU-4: 0.96% in Macro-F1 on IMDB to 15.61% in BLEU-4 | Improved |
| graph-aware instruction tuning of LLaMA3-8B | LLaMA3-8B · Freebase | Macro-F1: 3.76 -> 59.62 | Improved |
| graph-aware instruction tuning of LLaMA3-8B | LLaMA3-8B · WebNLG (500 instances) | BLEU-4: 36.54 -> 36.83 | Improved |
| MuseGraph versus TAPE (GCN) on homogeneous NC | MuseGraph (LLaMA3-8B) · Cora | Micro-F1: 88.19 vs 86.83 | Hurt |
| MuseGraph versus GNN baselines on link prediction | MuseGraph · link prediction | AUC: 17.54% | Improved |
Closest papers by meaning
- LoReC: Rethinking Large Language Models for Graph Data Analysis (NeurIPS 2026 · similarity 0.91)
- Large Language Model Meets Graph Neural Network in Knowledge Distillation (AAAI 2025 · similarity 0.91)
- Can Large Language Models Analyze Graphs like Professionals? A Benchmark, Datasets and Models (NeurIPS 2024 · similarity 0.90)
- How Do Large Language Models Understand Graph Patterns? A Benchmark for Graph Pattern Comprehension (ICLR 2025 · similarity 0.90)
- A Unified Graph Language Model for Multi-Domain Multi-Task Graph Alignment Instruction Tuning (NeurIPS 2026 · similarity 0.90)
- GOFA: A Generative One-For-All Model for Joint Graph Language Modeling (ICLR 2025 · similarity 0.89)
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.
