This is an example implementation of

GraLSP: Graph Neural Networks with Local Structural Patterns (AAAI 2020)

in TensorFlow.

The GraLSP learns representations for nodes in networks according to their local structural patterns as well its neighboring nodes.

Tensorflow 1.15.0rc0

Networkx 2.3

Python 3.7

I am not sure whether other versions are compatible, but I think as long as you use py3 and fix some issues in tensorflow, it will be fine.

If you want to use a custom dataset named X, please provide the following files and put them under the path data/X:

data/X/edges: formatted as a file where each line (a, b) describes a link from a to b. a and b are node ids which should be continuous in [0, num_nodes)

data/X/features.npy: A numpy array describing node features, where features[a] describes the feature vector for node id a in the data/X/edges file.

You can run the code in path ./ with the following script:

python main.py --model gralsp --dataset_name YOURDATASET

Of course there are other command line arguments which are listed in main.py.

We provide cora dataset as an example under data/cora. For other datasets like citeseer and aminer, you can find them via google and preprocess them according to the format.

Note that the data/cora/features.npy has undergone a dimensionality reduction via PCA, and is not identical to the original cora features.

@inproceedings{jin2020gralsp,
  title={GraLSP: Graph Neural Networks with Local Structural Patterns.},
  author={Jin, Yilun and Song, Guojie and Shi, Chuan},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  pages={4361--4368},
  year={2020}
}

If you use related codes or mention our work, you can cite us via the above bibtex.