Coupled-space Attack against Random-Walk-based Anomaly Detection

This is the source code for "Coupled-Space Attacks against Random-Walk-based Anomaly Detection".

Structure

.
├── Attack_random_walk_Bipartite
│   ├── Datasets
│   │   ├── Magazine_Subscriptions.csv
│   │   ├── __init__.py
│   │   ├── arxivData.json
│   │   └── paperCrawl.py
│   ├── DeepWalkAttack
│   │   ├── node_embedding_attack
│   │   │   ├── embedding.py
│   │   │   ├── main.py
│   │   │   ├── perturbation_attack.py
│   │   │   └── utils.py
│   │   └── setup.py
│   ├── __init__.py
│   ├── draw.py
│   ├── main.py
│   ├── models.py
│   ├── readme.md
│   ├── run.sh
│   └── utils.py
├── Attack_random_walk_Proximity
│   ├── 01_attack_graph
│   │   ├── DeepWalkAttack
│   │   │   ├── README.md
│   │   │   ├── node_embedding_attack
│   │   │   │   ├── __init__.py
│   │   │   │   ├── embedding.py
│   │   │   │   ├── main.py
│   │   │   │   ├── perturbation_attack.py
│   │   │   │   └── utils.py
│   │   │   └── setup.py
│   │   ├── baselines.py
│   │   ├── draw.py
│   │   ├── main.py
│   │   ├── model.py
│   │   ├── run.sh
│   │   └── utils.py
│   ├── 02_attack_feature
│   │   ├── draw.py
│   │   ├── main.py
│   │   ├── model.py
│   │   ├── run.sh
│   │   └── utils.py
│   └── DataSets
│       ├── Mnist
│       │   └── mnist.mat
│       └── NetworkIntrusion
│           └── KDD99
│               ├── data.pkl
│               ├── data_constraint_info.pkl
│               └── kddcup.data_10_percent.gz
├── README.md
└── requirements
    ├── py37.txt
    └── py37.yml

Environment Setup

Both of the two files are required:

conda env create -f requirements/py37.yml
conda activate py37
pip install -r requirements/py37.txt

if report: "ResolvePackageNotFound:xxx", or "No matching distribution found for xxx", just open the .yaml or .txt file and delete that line.

Dataset

KDD-99:

Please click the link below to download the "kddcup.data_10_percent.zip A 10% subset. (2.1M; 75M Uncompressed)" dataset:
https://www.kdd.org/kdd-cup/view/kdd-cup-1999/Data
The data preprocessing in main.py will generate KDD99/data.pkl and data_constraint_info.pkl files.

MNIST:

Please click the link below to download the "mnist.mat":
https://odds.cs.stonybrook.edu/mnist-dataset/

Author-Paper: This dataset is the papers crawled from arXiv preprint database.

The crawler is from the repository: GabrielePisciottaarxiv-dataset-download. There are many subjects of papers, such as cs.AI, math.ML. Nodes U are papers, and nodes V are authors, edges < u, v > means the author v shown in the paper u. Please see our paperCrawl.py and Attack_random_walk_Bipartite/main.py for detailed data processing.

Amazon Reviews: The Amason Reviews Data (2018).

This data includes different kinds of products and their ratings from users. Nodes U are products, and nodes V are users, edges < u, v > means the user v gives a review to the product u.

Our attack:

Attack_random_walk_Proximity (InDi-RWAD):
01_attack_graph: graph-space attack.
02_attack_graph: feature-space attack.

Attack_random_walk_Bipartite (Di-RWAD):
graph-space attack only.

Run our proposed attack examples:

python main.py -dataset "KDD99" -attack_mode 'closed-form' -gpuID 0
python main.py -dataset "KDD99" -attack_mode 'alternative' -gpuID 0

bash run.sh

Run the baselines:

examples:

python -u main.py -dataset "KDD99" -attack_mode 'DeepWalk' -gpuID 0
python -u main.py -dataset "KDD99" -attack_mode 'random' -gpuID 0

@misc{lai2023dualspace,
  title = {Coupled-Space Attacks against Random-Walk-based Anomaly Detection},
  author = {Yuni, Lai and Marcin, Waniek and Liying, Li and Jingwen, Wu and Yulin, Zhu
            and Tomasz P., Michalak and Talal, Rahwan and Kai, Zhou},
  year = {2023},
  journal = {arXiv preprint:2307.14387},
}

Yuni-Lai/CoupledAttackRW