This is the official implementation of the NeurIPS 2025 paper: TAMI: Taming Heterogeneity in Temporal Interactions for Temporal Graph Link Prediction.

Our code is developed based on the codebase of the paper: Towards Better Dynamic Graph Learning: New Architecture and Unified Library. We made modifications to four files: models/modules.py, train_link_prediction.py, evaluate_link_prediction.py, and evaluate_model_utils.py.

PyTorch 1.8.1, numpy, pandas, tqdm, and tabulate

Most of our experiments are conducted on 13 datasets from Towards Better Evaluation for Dynamic Link Prediction, including Wikipedia, Reddit, MOOC, LastFM, Enron, Social Evo., UCI, Flights, Can. Parl., US Legis., UN Trade, UN Vote, and Contact.

Nine popular link prediction methods for continuous-time dynamic graphs are included in our experiment: JODIE, DyRep, TGAT, TGN, CAWN, EdgeBank, TCL, GraphMixer, and DyGFormer.

If you want to load the best model configurations determined by the grid search, please set the load_best_configs argument to True.

• Example of training GraphMixer on Wikipedia dataset:

python train_link_prediction.py --dataset_name wikipedia --model_name GraphMixer --num_runs 5 --gpu 0

• If you want to use the best model configurations to train GraphMixer on Wikipedia dataset, run

python train_link_prediction.py --dataset_name wikipedia --model_name GraphMixer --load_best_configs --num_runs 5 --gpu 0

• Example of evaluating GraphMixer on the Wikipedia dataset:

python evaluate_link_prediction.py --dataset_name wikipedia --model_name GraphMixer --num_runs 5 --gpu 0

• If you want to use the best model configurations to evaluate GraphMixer on Wikipedia dataset, run

python evaluate_link_prediction.py --dataset_name wikipedia --model_name GraphMixer --load_best_configs --num_runs 5 --gpu 0

@inproceedings{yu2025tami,
  title={{TAMI}: Taming Heterogeneity in Temporal Interactions for Temporal Graph Link Prediction},
  author={Zhongyi Yu and Jianqiu Wu and Zhenghao Wu and Shuhan Zhong and Weifeng Su and Chul-Ho Lee and Weipeng Zhuo},
  booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems},
  year={2025}
}