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..
Simple and Efficient Heterogeneous Temporal Graph Neural Network
Authors: Yili Wang, Tairan Huang, Changlong He, Qiutong Li, Jianliang Gao
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that SE-HTGNN achieves up to 10 speed-up over the state-of-the-art and latest baseline while maintaining the best forecasting accuracy. |
| Researcher Affiliation | Academia | Yili Wang1,2, Tairan Huang1, Changlong He1, Qiutong Li1, Jianliang Gao1, 1Central South University 2The Hong Kong University of Science and Technology (Guangzhou) EMAIL, EMAIL |
| Pseudocode | No | No explicit pseudocode or algorithm block is present in the paper. The methodology is described through textual explanations and mathematical formulations in Section 4. |
| Open Source Code | Yes | All the code for reproducing the experiments is made available in the supplementary material accompanying the submission. |
| Open Datasets | Yes | We utilized two link prediction datasets: OGBN-MAG, Aminer, one node classification dataset YELP, and one node regression dataset COVID-19. We follow the splits provided by previous works [12, 9]... Aminer4: Aminer [9] is an academic citation dataset... OGBN-MAG5: The original OGBN-MAG dataset is a static heterogeneous network... Yelp6: Yelp [9] is a business review dataset... COVID-197: This dataset [12] contains both state and county level daily case reports... |
| Dataset Splits | Yes | Table 9: Statistics of datasets. Dataset Aminer... Data Split Training: 14 Validation: 1 Testing: 1. OGBN-MAG... Data Split Training: 8 Validation: 1 Testing: 1. YELP... Data Split Training: 10 Validation: 1 Testing: 1. COVID-19... Data Split Training: 244 Validation: 30 Testing: 30/60/90 |
| Hardware Specification | Yes | We provide the information on the computer resources in Appendix C.4: NVIDIA Ge Force RTX 3090 GPU 24GB memory, and Intel(R) Xeon(R) CPU E5-2650 v2 @ 2.60GHz CPU. |
| Software Dependencies | Yes | Software: Python 3.9.19, Deep Graph Library13 1.1.1 [53], Cuda 11.3, Py Torch14 1.12.1[54], Py Torch-Geometric15 2.5.3 [52]. |
| Experiment Setup | Yes | For all baselines and datasets, we use the default hyperparameters provided in the original source code, if available. Otherwise, we choose the number of messagepassing layers in {1, 2, 3} and the number of attention heads in {1, 2, 4, 8}. The hidden representation dimensionality is set as d = 64 except d = 8 for COVID-19. We record the best the best-performing results among them. Other hyperparameters for baselines are kept the same as in the original paper. The max number of epochs is 500, and we set the early stopping round on the validation set as 25 or 50 to alleviate over-fitting. We report the test performance based on the best epoch of the validation set. For our method, we adopt the Adam optimizer with a learning rate searched in {1,3,5} {10 2, 10 3}, and the weight decay rate is searched in {1,2,5} {10 4, 10 5}. The layer number of our methods is set as 2, except 3 for Aminer. |