Home

Awesome

VIKING

Code for paper accepted at PAKDD 2021, VIKING: Adversarial Attack on Network Embeddings via Supervised Network Poisoning Authors: Viresh Gupta, Tanmoy Chakraborty This repository has two main components:

  1. code - This folder contains the code for generating VIKING and VIKING^s perturbations. Additionally it also contains sample implementations of baselines and evaluation methods for node classification and link prediction tasks.

  2. data - This folder contains the three datasets used for reporting the given results.

Requirements for running the code

Make sure you have installed node2vec and LINE binaries and they are avalaible on path. (i.e node2vec and line should not print error on terminal). If you face errors while running line, make sure you have all required c-libraries on path. Also ensure pytorch and dgl library is installed.

Running instructions

From the root directory of this repository, call main.py as follows:

python code/simple_all.py cora lp

Since the results are quite long, it's ideal to save output to a file and later on tail the file for required output.

Acknowledgements

Some code has been borrowed from previous research work available at https://github.com/abojchevski/node_embedding_attack The same has been marked in the code files where applicable.