Adversarial Examples for Extreme Multilabel Text Classification [1]

The code is adapted from the source codes of BERT-ATTACK [2], APLC_XLNet [3], and AttentionXML [4]

Requirements

The code has been tested by the following packages:

• python==3.6.13
• boto3==1.17.70
• ruamel.yaml==0.16.12
• numpy==1.19.2
• scipy==1.5.4
• matplotlib==3.2.2
• scikit-learn==0.24.2
• transformers==2.9.0
• torch==1.4.0
• nltk==3.4
• pandas==1.1.5
• requests==2.25.1
• tqdm==4.60.0

A small experiment on Wikipedia-31K with only 10 samples per bin

Downolad the data and the APLC_XLNet model trained on this data as follows:

bash download_data_model.sh

For preprocessing the data, run the following:

bash preprocess.sh

To run positive-targeted attacks with 10 samples per bin, run the following:

bash pos_attack.sh

To check the results of the attacks, run the following:

bash results.sh

References

[1] Qaraei and Babbar, Adversarial Examples for Extreme Multilabel Text Classification, Arxiv 2021

[2] Li et al., BERT-ATTACK: Adversarial Attack Against BERT Using BERT, EMNLP 2020

[3] Ye et al., Pretrained Generalized Autoregressive Model with Adaptive Probabilistic Label Clusters for Extreme Multi-label Text Classification, ICML 2020

[4] You et al., AttentionXML: Label Tree-based Attention-Aware Deep Model for High-Performance Extreme Multi-Label Text Classification, NeurIPS 2019