tools-only / Evasion-Attack-against-LDA-Model

Code for the paper :

"EvaLDA: Efficient Evasion Attacks Towards Latent Dirichlet Allocation"

Authors : Qi Zhou*, Haipeng Chen*, Yitao Zheng, Zhen Wang. *Equal contribution

To appear in: AAAI, 2021.

Link : https://arxiv.org/abs/2012.04864

Instructions:

Installation

/Library/requirements.txt : the library for EvaLDA.

/Library/gensim.rar : the gensim which we modified.

Test data and pre-trained LDA model is in dataset/. EvaLDA.ipynb is the code.

Bert is needed, see https://github.com/hanxiao/bert-as-service/blob/master/README.md for more detail.

Run the code

1. Configure the environment according to the Library/requirements.txt.
2. Download Library/gensim.rar, unzip to the local python third-party library path, replace the original Gensim.
3. Download /dataset/ , you may need to manually modify the data and model loading in the code (EvaLDA.ipynb) according to the download path.
4. Download word2vec model: https://dl.fbaipublicfiles.com/fasttext/vectors-wiki/wiki.en.vec , put it in dataset/
5. Before run EvaLDA.ipynb, you should first open Bert server(see the bert link above).
6. Run the code.

Cite

@article{zhou2020evalda,
  title={EvaLDA: Efficient Evasion Attacks Towards Latent Dirichlet Allocation},
  author={Zhou, Qi and Chen, Haipeng and Zheng, Yitao and Wang, Zhen},
  journal={arXiv preprint arXiv:2012.04864},
  year={2020}
}