CLDG: Contrastive Learning on Dynamic Graphs (ICDE'23)
Code structure
| folder |
description |
| Data |
Datasets. |
| CLDG |
CLDG implementation code is provided. |
Datasets
| Dataset |
# nodes |
# temporal edges |
# classes |
| DBLP |
25,387 |
185,480 |
10 |
| Bitcoinotc |
5,881 |
35,592 |
3 |
| TAX |
27,097 |
315,478 |
12 |
| BITotc |
4,863 |
28,473 |
7 |
| BITalpha |
3,219 |
19,364 |
7 |
| TAX51 |
132,524 |
467,279 |
51 |
| Reddit |
898,194 |
2,575,464 |
3 |
Usage
# DBLP
python main.py --dataset dblp --hidden_dim 128 --n_classes 64 --n_layers 2 --fanout 20,20 --snapshots 4 --views 4 --strategy sequential --epochs 200 --GPU 0
# Bitcoinotc
python main.py --dataset bitcoinotc --hidden_dim 128 --n_classes 64 --n_layers 2 --fanout 10,10 --snapshots 4 --views 3 --strategy sequential --dataloader_size 64 --epochs 25 --GPU 0
# TAX
python main.py --dataset tax --hidden_dim 128 --n_classes 64 --n_layers 2 --fanout 20,20 --snapshots 4 --views 4 --strategy sequential --epochs 200 --GPU 0
# BITotc
python main.py --dataset bitotc --hidden_dim 128 --n_classes 64 --n_layers 2 --fanout 10,10 --snapshots 4 --views 4 --strategy random --epochs 50 --GPU 0
# BITalpha
python main.py --dataset bitalpha --hidden_dim 128 --n_classes 64 --n_layers 2 --fanout 20,20 --snapshots 6 --views 4 --strategy sequential --epochs 200 --GPU 0
# TAX51
python main.py --dataset tax51 --hidden_dim 128 --n_classes 64 --n_layers 2 --fanout 20,20 --snapshots 8 --views 5 --strategy random --epochs 200 --GPU 0
# reddit
python main.py --dataset reddit --hidden_dim 128 --n_classes 64 --n_layers 2 --fanout 20,20 --snapshots 5 --views 4 --strategy random --epochs 200 --GPU 0
Dependencies
- Python 3.7
- PyTorch 1.9.0+cu111
- dgl-cu110 0.7.1
Reference
If you find this repository useful in your research, please consider citing the following paper:
@inproceedings{xu2023cldg,
title={CLDG: Contrastive Learning on Dynamic Graphs},
author={Xu, Yiming and Shi, Bin and Ma, Teng and Dong, Bo and Zhou, Haoyi and Zheng, Qinghua},
booktitle={2023 IEEE 39th International Conference on Data Engineering (ICDE)},
pages={696--707},
year={2023},
organization={IEEE}
}