Deperturbation of Online Social Networks via Bayesian Label Transition

Authors: Jun Zhuang, Mohammad Al Hasan

Paper:

Accepted by SDM 2022

Abstract:

Online social networks (OSNs) classify users into different categories based on their online activities and interests, a task which is referred as a node classification task. Such a task can be solved effectively using Graph Convolutional Networks (GCNs). However, a small number of users, so-called perturbators, may perform random activities on an OSN, which significantly deteriorate the performance of a GCN-based node classification task. Existing works in this direction defend GCNs either by adversarial training or by identifying the attacker nodes followed by their removal. However, both of these approaches require that the attack patterns or attacker nodes be identified first, which is difficult in the scenario when the number of perturbator nodes is very small. In this work, we develop a GCN defense model, namely GraphLT, which uses the concept of label transition. GraphLT assumes that perturbators' random activities deteriorate GCN's performance. To overcome this issue, GraphLT subsequently uses a novel Bayesian label transition model, which takes GCN's predicted labels and applies label transitions by Gibbs-sampling-based inference and thus repairs GCN's prediction to achieve better node classification. Extensive experiments on seven benchmark datasets show that GraphLT considerably enhances the performance of the node classifier in an unperturbed environment; furthermore, it validates that GraphLT can successfully repair a GCN-based node classifier with superior performance than several competing methods.

Dataset:

kdd20_s1, kdd20_s2, cora, citeseer, amazoncobuy, coauthor, reddit
KDD20 Datasets can be downloaded from https://www.biendata.xyz/competition/kddcup_2020/

Getting Started:

Prerequisites

Linux or macOS
CPU or NVIDIA GPU + CUDA CuDNN
Python 3
pytorch, dgl, numpy, scipy, sklearn

Clone this repo

git clone https://github.com/junzhuang-code/GraphLT.git
cd GraphLT/graphLT

Install dependencies

For pip users, please type the command: pip install -r requirements.txt
For Conda users, you may create a new Conda environment using: conda env create -f environment.yml

Directories

graphLT:

1. pretrain.py: Train the GCN-based node classifier on given dataset
2. perturbation.py: Simulate non-malicious perturbations
3. main.py: Infer label by GraphLT
4. models_GCN.py/models_LT.py: Model scripts
5. load_data.py: Load data script
6. utils.py: Utils modules

data:

1. kdd20_s1/kdd20_s2: Two directories that store KDD20 datasets
2. noisy_label: A directory that stores labels

Runs

1. Train the GCN-based node classifier on given dataset
   python pretrain.py -data_name -model_name -NUM_EPOCHS -GPU -NOISY_RATIO -is_trainable
   e.g. python pretrain.py cora GCN 200 0 0.1 true

2. Simulate non-malicious perturbations and generate predicted labels
   python perturbation.py -data_name -model_name -NOISY_RATIO -TARGET_CLASS -SAMPLE_RATE -GPU
   e.g. python perturbation.py cora GCN 0.1 -1 1.0 0

3. Infer labels and evaluation
   python main.py -is_attacked -SAMPLE_RATE -GPU -is_inferable -NUM_INFER -WARMUP_STEP
   e.g. python main.py true 1.0 0 true 100 20

4. Visualization with Tensorboard
   Under the directory of "graphLT", run: Tensorboard --logdir=./runs/Logs_LT --port=8999

Cite

Please cite our paper if you think this repo is helpful.

@inproceedings{zhuang2022deperturbation,
  title={Deperturbation of Online Social Networks via Bayesian Label Transition},
  author={Zhuang, Jun and Al Hasan, Mohammad},
  booktitle={Proceedings of the 2022 SIAM International Conference on Data Mining (SDM)},
  pages={603--611},
  year={2022},
  organization={SIAM}
}