/SCAN-Pytorch

Primary LanguagePython

SCAN: Semi-supervisedly Co-embedding Attributed Networks

This repository contains the Python&Pytorch implementation for SCAN. Further details about SCAN can be found in this paper:

Zaiqiao Meng, Shangsong Liang, Jinyuan Fang, Teng Xiao. Semi-supervisedly Co-embedding Attributed Networks. (NeurIPS 2019)

The orignal tensorflow implementation for SCAN can be found in SCAN

Requirements

=================

  • Pytorch (1.0 or later)
  • python 3.6/3.7
  • scikit-learn
  • scipy

Run the demo

=================

python train.py

Result

The Link prediction performance AUC&AP score :

Dataset AUC AP
BLOGCATALOG 0.844 0.850
CORA 0.972 0.972
FLICKR 0.889 0.906

The Attribute inference performance AUC&AP score :

Dataset AUC AP
BLOGCATALOG 0.886 0.888
CORA 0.822 0.838
FLICKR 0.864 0.859

The node classification performance accuracy :

Dataset ACC of SCVA_SVM ACC of SCVA_DIS
BLOGCATALOG 0.834 0.844
CORA 0.736 0.822
FLICKR 0.695 0.800