Semi-supervised Intent Classification with GAN-BERT

This is the code implementation of TDS Article "Semi-supervised Intent Classification with GAN-BERT". The CLINC150 data is provided here. The codes used are based on the official repo of GAN-BERT, with some minor changes and additions. All of the changes in codes are listed here

Requirements

tensorflow-gpu==1.14.0

gast==0.2.2

git+https://github.com/guillaumegenthial/tf_metrics.git

Usage

First thing first you have to download the BERT model here. Then, if you have your own GPU you can download the Google Colab Notebook and run the script inside the notebook locally or you can upload this repo to your Google Drive then run it using the Google Colab's GPU.

References

GAN-BERT Paper: https://www.aclweb.org/anthology/2020.acl-main.191.pdf

CLINC150 Paper: https://www.aclweb.org/anthology/D19-1131.pdf