Home

Awesome

HeGAN

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

Evironment Setting

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

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

The format of input pre-trained data

The format of output embedding

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},

}