Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/jind11/TextFooler
A Model for Natural Language Attack on Text Classification and Inference
https://github.com/jind11/TextFooler
adversarial-attacks bert bert-model natural-language-inference natural-language-processing text-classification
Last synced: 4 days ago
JSON representation
A Model for Natural Language Attack on Text Classification and Inference
- Host: GitHub
- URL: https://github.com/jind11/TextFooler
- Owner: jind11
- License: mit
- Created: 2019-09-03T20:06:21.000Z (about 5 years ago)
- Default Branch: master
- Last Pushed: 2022-12-08T10:46:49.000Z (almost 2 years ago)
- Last Synced: 2024-08-01T23:32:02.198Z (3 months ago)
- Topics: adversarial-attacks, bert, bert-model, natural-language-inference, natural-language-processing, text-classification
- Language: Python
- Size: 2.77 MB
- Stars: 481
- Watchers: 15
- Forks: 79
- Open Issues: 23
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- awesome-MLSecOps - TextFooler
README
# TextFooler
A Model for Natural Language Attack on Text Classification and Inference
This is the source code for the paper: [Jin, Di, et al. "Is BERT Really Robust? Natural Language Attack on Text Classification and Entailment." arXiv preprint arXiv:1907.11932 (2019)](https://arxiv.org/pdf/1907.11932.pdf). If you use the code, please cite the paper:
```
@article{jin2019bert,
title={Is BERT Really Robust? Natural Language Attack on Text Classification and Entailment},
author={Jin, Di and Jin, Zhijing and Zhou, Joey Tianyi and Szolovits, Peter},
journal={arXiv preprint arXiv:1907.11932},
year={2019}
}
```
## Data
Our 7 datasets are [here](https://bit.ly/nlp_adv_data).
## Prerequisites:
Required packages are listed in the requirements.txt file:
```
pip install -r requirements.txt
```
## How to use
* Run the following code to install the **esim** package:
```
cd ESIM
python setup.py install
cd ..
```
* (Optional) Run the following code to pre-compute the cosine similarity scores between word pairs based on the [counter-fitting word embeddings](https://drive.google.com/open?id=1bayGomljWb6HeYDMTDKXrh0HackKtSlx).
```
python comp_cos_sim_mat.py [PATH_TO_COUNTER_FITTING_WORD_EMBEDDINGS]
```
* Run the following code to generate the adversaries for text classification:
```
python attack_classification.py
```
For Natural langauge inference:
```
python attack_nli.py
```
Examples of run code for these two files are in [run_attack_classification.py](https://github.com/jind11/TextFooler/blob/master/run_attack_classification.py) and [run_attack_nli.py](https://github.com/jind11/TextFooler/blob/master/run_attack_nli.py). Here we explain each required argument in details:
* --dataset_path: The path to the dataset. We put the 1000 examples for each dataset we used in the paper in the folder [data](https://github.com/jind11/TextFooler/tree/master/data).
* --target_model: Name of the target model such as ''bert''.
* --target_model_path: The path to the trained parameters of the target model. For ease of replication, we shared the [trained BERT model parameters](https://drive.google.com/drive/folders/1wKjelHFcqsT3GgA7LzWmoaAHcUkP4c7B?usp=sharing), the [trained LSTM model parameters](https://drive.google.com/drive/folders/108myH_HHtBJX8MvhBQuvTGb-kGOce5M2?usp=sharing), and the [trained CNN model parameters](https://drive.google.com/drive/folders/1Ifowzfers0m1Aw2vE8O7SMifHUhkTEjh?usp=sharing) on each dataset we used.
* --counter_fitting_embeddings_path: The path to the counter-fitting word embeddings.
* --counter_fitting_cos_sim_path: This is optional. If given, then the pre-computed cosine similarity scores based on the counter-fitting word embeddings will be loaded to save time. If not, it will be calculated.
* --USE_cache_path: The path to save the USE model file (Downloading is automatic if this path is empty).
Two more things to share with you:
1. In case someone wants to replicate our experiments for training the target models, we shared the used [seven datasets](https://drive.google.com/open?id=1N-FYUa5XN8qDs4SgttQQnrkeTXXAXjTv) we have processed for you!
2. In case someone may want to use our generated adversary results towards the benchmark data directly, [here it is](https://drive.google.com/drive/folders/12yeqcqZiEWuncC5zhSUmKBC3GLFiCEaN?usp=sharing).