Awesome

HeGAN

Source code for paper "Adversarial Learning on Heterogeneous Information Network (KDD2019)"

Evironment Setting

• Python == 2.7.3

• Tensorflow == 1.12.0

• Numpy == 1.15.1

Parameter Setting (see config.py)

batch_size : The size of batch.

lambda_gen, lambda_dis : The regularization for generator and discriminator, respectively.

lr_gen, lr_dis : The learning rate for generator and discriminator, respectively.

n_epoch : The maximum training epoch.

sig : The variance of gaussian distribution in generator.

g_epoch, d_epoch: The number of generator and discriminator training per epoch.

n_sample : The size of sample

n_emb : The embedding size

Files in the folder

• data/: The training data

• results/: The learned embeddings of generator ane discriminator.

• code/: The source codes

• pre_train/: The pre-trained node embeddings (Note: The dimension of pre-trained node embeddings should equal n_emb)

Data

We provide three datasets: DBLP, Yelp and Aminer, The detailed description of the three datasets can refer to https://github.com/librahu/Heterogeneous-Information-Network-Datasets-for-Recommendation-and-Network-Embedding

The format of input training data

• Each line: source_node target_node relation

The format of input pre-trained data

• The first line: node_num embedding_dim

• Each line : node_id embdeeing_1 embedding_2, ...

The format of output embedding

• The first line: node_num embedding_dim

• Each line : node_id embdeeing_1 embedding_2, ...

Basic Usage

cd code

python he_gan.py

Reference

@inproceedings{

author = {Binbin Hu, Yuan Fang and Chuan Shi.},

title = {Adversarial Learning on Heterogeneous Information Network},

booktitle = {Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining},

year = {2019},

publisher = {ACM},

address = {Anchorage, Alaska, USA},

year = {2019},

keywords = {Heterogeneous Information Network, Network Embedding, Generative Adversarial Network},

}