1. Overview

PyTorch Implementation for "Discrete-time Temporal Network Embedding via Implicit Hierarchical Learning in Hyperbolic Space (KDD2021)"

Authors: Menglin Yang; Min Zhou; Marcus Kalander; Zengfeng Huang; Irwin King

Paper: https://arxiv.org/pdf/2107.03767.pdf

Code: https://github.com/marlin-codes/HTGN

2. Setup

2.1 Environment

pip install -r requirements.txt

2.2 Datasets

The data is cached in ./data/input/cached. The raw datasets are in the folder ./data/input/raw

Note:

• The dataset with name dblp is the COLAB dataset in the paper.
• Link: to download more datasets

3. Examples

3.0 Go to the script at first

cd ./script

3.1 To quickly run the code:

python main.py --model=HTGN --dataset=enron10

3.2 To run more examples about HTGN, static method, run the following

bash example/run_htgn.sh
bash example/run_static.sh
bash example/run_grugcn.sh

3.3 To run DySAT, please refer to DySAT;

3.4 To run EvolveGCN, please refer to EvolveGCN or run the file bash example/run_evolvegcn.sh;

3.5 To run VGRNN, please refer to VGRNN;

Note: for a unified comparison, we use inner product as decoder for the baseline models.

4. Some of the code was forked from the following repositories

• HGCN
• HyperRNN
• EvolveGCN
• DySAT
• VGRNN
• CTGCN
• GloDyNE

5. Reference

[1] Discrete-time Temporal Network Embedding via Implicit Hierarchical Learning in Hyperbolic Space.

[2] Evolvegcn: Evolving graph convolutional networks for dynamic graphs.

[3] Variational graph recurrent neural networks.

[4] DySAT: Deep neural representation learning on dynamic graphs via self-attention networks

6. Citation

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

@inproceedings{yang2021discrete,
  title={Discrete-time Temporal Network Embedding via Implicit Hierarchical Learning in Hyperbolic Space},
  author={Yang, Menglin and Zhou, Min and Kalander, Marcus and Huang, Zengfeng and King, Irwin},
  booktitle={Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery \& Data Mining},
  pages={1975--1985},
  year={2021}
}