CAMERA: Adapting to Semantic Camouflage in Unsupervised Text Attributed Graph Fraud Detection
May 19, 2026 ยท View on GitHub
This is the official implementation of the following paper: CAMERA: Adapting to Semantic Camouflage in Unsupervised Text Attributed Graph Fraud Detection, Accepted by IJCAI 2026
CAMERA employs a mixture of expert architectures to disentangle structural, statistical, and semantic signals, enabling robust anomaly detection under distribution shift without relying on labeled data.
Environment
Python 3.11
- torch==2.5.1+cu121
- torch-sparse==0.6.18+pt25cu121
- numpy==1.26.4
- scikit-learn==1.6.1
- pickle
- sentence-transformers==3.3.1
- torch-geometric==2.6.1
Usage
Step 0: Set up the dataset
You can download the dataset points from: https://drive.google.com/file/d/1z75C7Cul7z4SYNhyZn6Jit7cY18xFbEA/view?usp=sharing
To make experimentation easier and reduce reproduction costs, we also provide text encodings generated using the OpenAI encoding model, available at: https://drive.google.com/file/d/1PqaIm_m3DVZbjdoe8NIuMupik3rtJO42/view?usp=sharing
Step 1 Text Encoding
First, generate text embeddings for each dataset.
python text_encoder.py
Step 2 Run CAMERA
After text embeddings are prepared, run the main training and evaluation script with dataset specific hyperparameters.
Below are the hyperparameter settings used in our experiments.
python run.py --dataset_name reddit --epochs 1200 --alpha 5.0 --beta 0.1 --lr 1e-3
python run.py --dataset_name instagram --epochs 15 --alpha 10.0 --beta 10.0 --lr 5e-5
Amazon
python run.py --dataset_name amazon --epochs 15 --alpha 0.1 --beta 1.0 --lr 5e-5
YelpChi
python run.py --dataset_name yelpchi --epochs 450 --alpha 0.1 --beta 10.0 --lr 1e-3
Citation
If you find our repository useful for your research, please consider citing our paper:
@inproceedings{pan2026CAMERA,
title={CAMERA: Adapting to Semantic Camouflage in Unsupervised Text-Attributed Graph Fraud Detection},
author={Pan, Junjun and Liu, Yixin and Zheng, Yu and Chi, Lianhua and Liew, Alan Wee-Chung and Pan, Shirui},
booktitle={International Joint Conference on Artificial Intelligence},
year={2026}
}