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; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
ODNet: Opinion Dynamics-Inspired Neural Message Passing for Graphs and Hypergraphs
Authors: Bingxin Zhou, Outongyi Lv, Jing Wang, Xiang Xiao, Weishu Zhao
TMLR 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We validate the performance of ODNet through classic node-level representation learning tasks on a variety of social networks described by homophilic/heterophilic graphs or hypergraphs. ... Result Analysis Tables 1-2 present the average accuracy for predicting node labels in homophilic and heterophilic graphs, respectively. ODNet consistently ranks among the top-performing methods with minimal variance. |
| Researcher Affiliation | Academia | Bingxin Zhou EMAIL Institute of Natural Sciences Shanghai Jiao Tong University; Outongyi Lv EMAIL Institute of Natural Sciences Shanghai Jiao Tong University; Jing Wang EMAIL School of Oceanography Shanghai Jiao Tong University; Xiang Xiao EMAIL School of Life Sciences and Biotechnology Shanghai Jiao Tong University; Weishu Zhao EMAIL School of Life Sciences and Biotechnology Shanghai Jiao Tong University |
| Pseudocode | No | The paper describes mathematical formulations and update rules for ODNet, but it does not include a distinct pseudocode or algorithm block. |
| Open Source Code | No | All implementations are programmed with Py Torch-Geometric (version 2.0.1) (Fey & Lenssen, 2019) and Py Torch (version 1.7.0) and run on NVIDIAr Tesla A100 GPU with 6, 912 CUDA cores and 80GB HBM2 mounted on an HPC cluster. All the details to reproduce our results have been included in the submission. The program will be publicly available upon acceptance. |
| Open Datasets | Yes | We encompass seven homophilic graphs (Cora (Mc Callum et al., 2000), Citeseer (Sen et al., 2008), Pubmed (Namata et al., 2012), Coauthor CS (Shchur et al., 2018), Computer (Namata et al., 2012), Photo (Namata et al., 2012), and ogb-ar Xiv Hu et al. (2020)), three heterophilic graphs (Texas, Wisconsin, and Cornell from the Web KB dataset (García-Plaza et al., 2016)). ... The hypergraph variant of ODNet undergoes an evaluation through semisupervised node classification tasks conducted on four benchmark hypergraphs extracted from Yadati et al. (2019). |
| Dataset Splits | Yes | For homophilic datasets, we utilized 10 random weight initializations and random splits, with each combination randomly selecting 20 instances from each class. In heterophilic and hypergraph datasets, we used the fixed 10 training/validation splits in Pei et al. (2020) and Yadati et al. (2019). |
| Hardware Specification | Yes | All implementations are programmed with Py Torch-Geometric (version 2.0.1) (Fey & Lenssen, 2019) and Py Torch (version 1.7.0) and run on NVIDIAr Tesla A100 GPU with 6, 912 CUDA cores and 80GB HBM2 mounted on an HPC cluster. |
| Software Dependencies | Yes | All implementations are programmed with Py Torch-Geometric (version 2.0.1) (Fey & Lenssen, 2019) and Py Torch (version 1.7.0) and run on NVIDIAr Tesla A100 GPU with 6, 912 CUDA cores and 80GB HBM2 mounted on an HPC cluster. |
| Experiment Setup | Yes | For ODNet, we trained the model using a neural ODE solver with Dormand Prince adaptive step size scheme (DOPRI5). For homophilic datasets, we utilized 10 random weight initializations and random splits, with each combination randomly selecting 20 instances from each class. In heterophilic and hypergraph datasets, we used the fixed 10 training/validation splits in Pei et al. (2020) and Yadati et al. (2019). |