Please cite the following work if you find the code useful.
@inproceedings{yang2018node,
Author = {Yang, Carl and Liu, Mengxiong and Zheng, Vincent and Han, Jiawei},
Booktitle = {ASONAM},
Title = {Node, motif and subgraph: learning network functional blocks through structural convolution},
Year = {2018}
}
Contact: Carl Yang (yangji9181@gmail.com)
pip install dill tqdm tensorflow
- Match instances with motifs
# for cascade task
python preprocess.py
# for classification task
python prepare.py
- Training and evaluating
python main.py
- To change dataset, modify the data_dir parameter in flags in main.py
- kernel.json under each dataset directory defines the motifs to be matched, modify it to customize the motifs
- For details of hyper-parameters, please refer to the comment in flags in main.py
- graph.txt contains the edge list of the complete graph, graph is undirected
- train.txt contains the training data, each line is a data point, each data point is a subgraph
- train/subgraph/ contains all the data points, one data point per file, each represented as an edge list
- train/meta/ contains all the matched instances of motifs, one data point per file