OpenHGNN/openhgnn/output/HeGAN

• HeGAN[KDD2019]
• • How to run
  • Performance: NodeClassification
  • Dataset
  • • Description
    • Entity Statistics
  • TrainerFlow: node_classification
  • • • model
  • More
  • • • Contributor
      • If you have any questions,

HeGAN[KDD2019]

• paper: Adversarial Learning on Heterogeneous Information Networks
• Code from author: HeGAN

How to run

• Clone the Openhgnn-DGL

  python main.py -m HeGAN -d yelp4HeGAN -t node_classification -g -1
  

  If you do not have gpu, set -gpu -1.

Performance: NodeClassification

dataset	paper's	author's	ours
yelp(%)	85.24	80.01	79.35

Dataset

• We process the yelp dataset given by yelp. It saved as dgl.heterograph and can be loaded by dgl.load_graphs

Description

• Yelp
  Dataset Link

  Entity Statistics

  Entity	#Entity
  User	1,286
  Business	2,614
  Service	2
  Star level	9
  Reservation	2
  Category	3

TrainerFlow: node_classification

model

• HeGAN

• 	HeGAN is a wrapper of Generator and Discriminator.
  

• Generator

• A Discriminator D eveluates the connectivity between the pair of nodes u and v w.r.t. a relation r.

• Discriminator

• A generator G samples fake node embeddings from a continuous distribution. The distribution is Gaussian distribution.

More

Contributor

Hui Han [GAMMA LAB]

If you have any questions,

Submit an issue or email to hanhui@bupt.edu.cn