Research map / Graph learning / Graph self-supervised learning
Glance for Context: Learning When to Leverage LLMs for Node-Aware GNN-LLM Fusion
Donald Loveland, Yao-An Yang, Danai Koutra
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 |
|---|---|---|---|
| Route nodes to the LLM with a learned router instead of using GNN or LLM-enhanced GNN uniformly | GNN backbones · Arxiv23 | accuracy: 80.2 -> 82.1 | Improved |
| Add GLANCE routing on heterophilous nodes (h<0.25) | GNN with LLM enhancement · Cora | accuracy by homophily bin: 32.0 -> 46.4 | Improved |
| Add GLANCE on Pubmed overall | LLM-enhanced GNNs · Pubmed | accuracy: 92.2 -> 92.6 | Improved |
| Use uncertainty routing heuristic across datasets | GCN enhanced with LLM routing · Cora vs Pubmed | net correction score: Cora -0.09; Pubmed 0.20 | Mixed |
| Route by estimated local homophily instead of random | GCN enhanced · Cora | net correction score: -0.02 -> -0.07 | Hurt |
| Route by estimated local homophily instead of random | GCN enhanced · Pubmed | net correction score: 0.06 -> 0.20 | Improved |
| Use LOGIN-filtered graphs instead of LLM-enhanced features | GCN · Arxiv23 | accuracy: 76.2 -> 65.7 | Hurt |
| Increase routing budget K from 8 to 12 to 16 (ablation) | GLANCE · Pubmed and Arxiv23, h_v<0.25 | accuracy: +3.4% then +3.0% | Improved |
| Increase K from 8 to 12 (ablation) | GLANCE · Cora, h_v<0.25 | accuracy | Hurt |
| Add the homophily routing feature (vs ablating it) (ablation) | GLANCE · Cora, h_v<0.5 | accuracy: -6.5% when removed | Improved |
| Add all routing features (vs disabling each one) (ablation) | GLANCE · Cora, Pubmed, Arxiv23 | accuracy: -0.38%, -1.07%, -0.65% average drop | Improved |
| Apply GLANCE to large graph at ~6.25% query rate | GCNII enhanced · Arxiv-Year | accuracy: 49.6 -> 49.8 | Improved |
Closest papers by meaning
- Label-free Node Classification on Graphs with Large Language Models (LLMs) (ICLR 2024 · similarity 0.91)
- Attention Mechanisms Perspective: Exploring LLM Processing of Graph-Structured Data (ICML 2025 · similarity 0.91)
- LoReC: Rethinking Large Language Models for Graph Data Analysis (NeurIPS 2026 · similarity 0.91)
- GraphRouter: A Graph-based Router for LLM Selections (ICLR 2025 · similarity 0.90)
- LLM Enhancers for GNNs: An Analysis from the Perspective of Causal Mechanism Identification (ICML 2025 · similarity 0.89)
- LLaGA: Large Language and Graph Assistant (ICML 2024 · 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.
