GMASK

Code for the paper "Explaining Neural Network Predictions on Sentence Pairs via Learning Word-Group Masks"

Requirement:

• python == 3.6.11
• pytorch == 1.4.0
• numpy == 1.18.5

Data:

Download the data and put it in the same folder with the code.

Train models on different datasets:

Train the decomposable attention model (DAttn) and BERT model on different datasets by running

python train.py

Note that the code is for the e-SNLI dataset. For the BERT model on other datasets, set --task_name with the data name quora/qqp/mrpc. For the MRPC dataset, set --max_seq_length as 100.

For the DAttn model on other datasets, utilize the corresponding DataLoader and Sampler by revising lines 1, 2, 6 in load_data.py. Set --data_path as train.tsv. Set the output dimension of the final linear layer of the DAttn model as 2 (line 59 in deatten_model.py).

Explain models on test data via GMASK:

Explain the well-trained model by running

python explain.py

For each test example, we save the words and their indexes in the order of importance as the explanation.

Acknowledgments

The code was built on

• https://github.com/huggingface/transformers
• https://github.com/asappresearch/rationale-alignment
• https://github.com/libowen2121/SNLI-decomposable-attention

Reference:

If you find this repository helpful, please cite our paper:

@inproceedings{chen-etal-2021-explaining,
    title = "Explaining Neural Network Predictions on Sentence Pairs via Learning Word-Group Masks",
    author = "Chen, Hanjie  and
      Feng, Song  and
      Ganhotra, Jatin  and
      Wan, Hui  and
      Gunasekara, Chulaka  and
      Joshi, Sachindra  and
      Ji, Yangfeng",
    booktitle = "Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
    month = jun,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2021.naacl-main.306",
    pages = "3917--3930",
    abstract = "Explaining neural network models is important for increasing their trustworthiness in real-world applications. Most existing methods generate post-hoc explanations for neural network models by identifying individual feature attributions or detecting interactions between adjacent features. However, for models with text pairs as inputs (e.g., paraphrase identification), existing methods are not sufficient to capture feature interactions between two texts and their simple extension of computing all word-pair interactions between two texts is computationally inefficient. In this work, we propose the Group Mask (GMASK) method to implicitly detect word correlations by grouping correlated words from the input text pair together and measure their contribution to the corresponding NLP tasks as a whole. The proposed method is evaluated with two different model architectures (decomposable attention model and BERT) across four datasets, including natural language inference and paraphrase identification tasks. Experiments show the effectiveness of GMASK in providing faithful explanations to these models.",
}