Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
GraphTOP: Graph Topology-Oriented Prompting for Graph Neural Networks
Authors: Xingbo Fu, Zhenyu Lei, Zihan Chen, Binchi Zhang, Chuxu Zhang, Jundong Li
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on five graph datasets under four pre-training strategies demonstrate that our proposed Graph TOP outshines six baselines on multiple node classification datasets. |
| Researcher Affiliation | Academia | Xingbo Fu University of Virginia Charlottesville, VA, USA EMAIL Zhenyu Lei University of Virginia Charlottesville, VA, USA EMAIL Zihan Chen University of Virginia Charlottesville, VA, USA EMAIL Binchi Zhang University of Virginia Charlottesville, VA, USA EMAIL Chuxu Zhang University of Connecticut Storrs, CT, USA EMAIL Jundong Li University of Virginia Charlottesville, VA, USA EMAIL |
| Pseudocode | Yes | The overall algorithm of Graph TOP is provided in Algorithm 1. |
| Open Source Code | Yes | Our code is available at https://github.com/xbfu/Graph TOP. |
| Open Datasets | Yes | We adopt five real-world graph datasets from various domains to evaluate the performance of our framework. These datasets include Cora [47], Pub Med [47], Amazon [31], Minesweeper [31], and Flickr [53]. |
| Dataset Splits | Yes | All the experimental results are based on the 5-shot setting. |
| Hardware Specification | Yes | We run our experiments using a server equipped with 512 GB of memory, 128 AMD EPYC 7543 32-core CPUs, and 6 NVIDIA RTX A6000 GPUs, each of which has 48 GB of memory. |
| Software Dependencies | No | The paper mentions using the Adam optimizer and a 2-layer GCN model but does not specify software dependencies like Python, PyTorch, or TensorFlow with their respective version numbers. |
| Experiment Setup | Yes | We use a 2-layer GCN [23] as the GNN model for each graph prompting method. The hidden size is 128. All the experimental results are based on the 5-shot setting. Each method is trained using the Adam optimizer [22] with a learning rate of 0.005. The number of epochs is set to 500 for graph prompting. We set γ = 0.5 in our experiments. We conduct a grid search for λ1 and λ2. The reported performance is the average result of five runs with different random seeds. |