PyTorch Implementation for "HGWaveNet: A Hyperbolic Graph Neural Network for Temporal Link Prediction (WWW2023)“

Authors: Qijie Bai, Changli Nie, Haiwei Zhang, Dongming Zhao, Xiaojie Yuan

Run python main.py --dataset=dblp for example.

Some critical config parameters:

• --dataset, --data_pt_path: name and parent path of dataset
• --test_length: number of snapshots for test set
• --spatial_dilated_factors: a list, dilated factor for HDGC module
• --casual_conv_depth, --casual_conv_kernel_size: number of temporal casual convolution layers, temporal casual convolution kernel size, used to config the receival field of HDCC module

For all config parameter description, please refer to ./cofig.py

The demo dataset can be found in ./data/

The input data is a serialized dict object by torch.save(). It has the following keys:

• edge_index_list: a list of torch tensors, each of which is the edge index of single snapshot
• pedges, nedges: a list of torch tensors, each of which is the sampled positive and negative edge index of single snapshot for temporal link prediction
• new_pedges, new_nedges: a list of torch tensors, each of which is the sampled positive and negative edge index of single snapshot for temporal new link prediction
• num_nodes: number of nodes of the whole temporal graph
• time_length: number of snapshots, numerically equal to the length of edge_index_list, pedges, nedges, new_pedges and new_nedges
• weights: input node feature tensor, remain None if there is no input feature

If you find this code useful, please cite the following paper:

@inproceedings{bai2023hgwavenet,
  title={HGWaveNet: A Hyperbolic Graph Neural Network for Temporal Link Prediction},
  author={Bai, Qijie and Nie, Changli and Zhang, Haiwei and Zhao, Dongming and Yuan, Xiaojie},
  booktitle={Proceedings of the ACM Web Conference 2023},
  pages={523--532},
  year={2023}
}

**Note: ** there is a typo in our code that we misspelled causal instead of casual, but it does not affect the code running results.