Hierarchical Weighted Self-contrastive Learning

Data and code for paper titled Fine-grained Category Discovery under Coarse-grained supervision with Hierarchical Weighted Self-contrastive Learning (EMNLP 2022 Long paper)

Fine-grained Category Discovery under Coarse-grained supervision (FCDC) aims to discover novel fine-grained categories automatically based on the coarse-grained labeled data which are easier and cheaper to obtain.

Contents

1. Data

2. Model

3. Requirements

4. Running

5. Results

6. Thanks

7. Citation

Data

We performed experiments on three public datasets: clinc, wos and hwu64, which have been included in our repository in the data folder ' ./data '.

Model

Our model mainly contains three components: BERT, Dynamic Queue and Momentum BERT.

Requirements

• python==3.8
• pytorch==1.10.0
• transformers==4.19.2
• scipy==1.8.0
• numpy==1.21.6
• scikit-learn==1.1.1

Running

Training and testing our model through the bash scripts:

sh scripts/run.sh

You can also add or change parameters in run.sh. (More parameters are listed in init_parameter.py)

Results

It should be noted that the experimental results may be slightly different because of the randomness of clustering when testing.

Thanks

Some code references the following repositories:

• DeepAligned
• SimCSE
• MoCo
• SupContrast

Citation

If our paper or code is helpful to you, please consider citing our paper:

@inproceedings{an-etal-2022-fine,
    title = "Fine-grained Category Discovery under Coarse-grained supervision with Hierarchical Weighted Self-contrastive Learning",
    author = "An, Wenbin  and
      Tian, Feng  and
      Chen, Ping  and
      Tang, Siliang  and
      Zheng, Qinghua  and
      Wang, QianYing",
    booktitle = "Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing",
    year = "2022",
    address = "Abu Dhabi, United Arab Emirates",
    publisher = "Association for Computational Linguistics",
    pages = "1314--1323",
}