Research map / Graph learning / GNN architectures
HGPrompt: Bridging Homogeneous and Heterogeneous Graphs for Few-Shot Prompt Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes HGPrompt, a dual-template, dual-prompt framework that unifies link-prediction pre-training and few-shot downstream tasks for both homogeneous and heterogeneous graphs.
Problem
Few-shot node classification on graphs (homogeneous and heterogeneous) where self-supervised pre-training and downstream tasks mismatch in format and graph heterogeneity, limiting prompt-based transfer.
Approach
Unifies pre-training and downstream tasks via a dual-template (task format and graph heterogeneity) and adapts with dual prompts (feature prompt and heterogeneity/meta-structure prompt), extending GraphPrompt to heterogeneous graphs.
Key results
- On OAG node classification, HGPrompt+ achieves 29.82% MicroF1 and 28.02% MacroF1, versus 25.27% MicroF1 for GraphPrompt.
- HGPrompt outperforms its Variant 3 (per-type GraphPrompt aggregation) and HGPrompt+ surpasses Variant 3+, per the authors' rebuttal ablation.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use HGPrompt instead of GraphPrompt | OAG node classification · OAG | MicroF (%): 25.27±1.59 -> 27.39±1.40 | Helped |
| Use HGPrompt+ instead of GraphPrompt | OAG node classification · OAG | MicroF (%): 25.27±1.59 -> 29.82±1.56 | Helped |
| Use HGPrompt instead of GraphPrompt | OAG node classification · OAG | MarcoF (%): 23.24±1.47 -> 26.19±1.65 | Helped |
| Use HGPrompt+ instead of GraphPrompt | OAG node classification · OAG | MarcoF (%): 23.24±1.47 -> 28.02±1.71 | Helped |
| Add a graph template (ablation) | HGPrompt Variant 2 · Tab. 3 benchmark settings | performance | Helped |
| Add a graph template (ablation) | HGPrompt+ Variant 2+ · Tab. 3 benchmark settings | performance | Helped |
Limitations
- Focus is on node classification in few-shot settings; pre-training task choice (link prediction) is not extensively explored in the text.
- Baseline backbones selected per baseline for best performance, which complicates fair comparison.
Experimental setup
Datasets: OAG Simple-HGN benchmark heterogeneous graph datasets
Benchmarks: OAG
Metrics: MicroF1 MacroF1
Baselines: GraphPrompt HeCo GPPT DGI GraphCL Simple-HGN
Builds on: GraphPrompt GPPT Simple-HGN Deep graph infomax
Closest papers by meaning
- HeterGP: Bridging Heterogeneity in Graph Neural Networks with Multi-View Prompting (AAAI 2025 · similarity 0.92)
- HePa: Heterogeneous Graph Prompting for All-Level Classification Tasks (AAAI 2025 · similarity 0.92)
- Edge Prompt Tuning for Graph Neural Networks (ICLR 2025 · similarity 0.91)
- HG-Adapter: Improving Pre-Trained Heterogeneous Graph Neural Networks with Dual Adapters (ICLR 2025 · similarity 0.91)
- One Prompt Fits All: Universal Graph Adaptation for Pretrained Models (NeurIPS 2025 · similarity 0.90)
- HGOT: Self-supervised Heterogeneous Graph Neural Network with Optimal Transport (ICML 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.