• TIP-GNN
  • Setup the environment
  • Data Preprocess
  • Run the Script
    • Temporal Link Prediction (Transductive)
    • Temporal Link Prediction (Inductive)
    • Temporal Node Classification
  • Cite us

Authors: Tongya Zheng, Zunlei Feng, Tianli Zhang, Yunzhi Hao, Mingli Song, Xingen Wang, Xinyu Wang, Ji Zhao, Chun Chen

Code for Transition Propagation Graph Neural Networks for Temporal Networks.

• conda create -n tip python=3.9 -y

• pip install -r requirements.txt

• My torch version is torch-1.10.2+cu113

We have preprocessed most temporal graphs in the data/format_data directory, and placed the JODIE datasets at Google drive, which can be downloaded and placed at the data/format_data.

bash init.sh

We use init.sh to make necessary directories for our experiments to store generated datasets by data/*, boost the training speed by gumbel_cache and sample_cache, record training details by log, record testing results by results and nc-results, save our trained models by ckpt and saved_models.

python data_unify.py -t datasplit
python data_unify.py -t datalabel

We use -t datasplit to split datasets into the training, validation and testing set according to the ratios.

In the setting of transductive temporal link prediction, we use trainable node embeddings.

• python exper_edge_np.py -d fb-forum

In the setting of inductive temporal link prediction,, we firstly generate features for each node to perform inductive link prediction.

• python inductive_util.py -d fb-forum

• python inductive_edge_np.py -d fb-forum

In the setting of temporal node classification prediction, we use the edge features and freeze the node embeddings as all zeros.

Firstly, we have to train a pre-trained link prediction model following TGAT.

• python exper_edge_np.py -d JODIE-wikipedia -t node -f

• python exper_node_np.py -d JODIE-wikipedia -f --balance --binary

@article{zheng2022transition,
  title={Transition Propagation Graph Neural Networks for Temporal Networks},
  author={Zheng, Tongya and Feng, Zunlei and Zhang, Tianli and Hao, Yunzhi and Song, Mingli and Wang, Xingen and Wang, Xinyu and Zhao, Ji and Chen, Chun},
  journal={IEEE Transactions on Neural Networks and Learning Systems},
  year={2022},
  pages={1--13},
  publisher={IEEE}
}