A novel approach for BERT usage in an adversarial unsupervised domain adaptation manner for a NLP tasks. The topic is Unsupervised domain adaptation between two Amazon product reviews categories with BERT and a domain discriminator network for the sentiment analysis.
Please cite original BERT paper when using the code.
The code based on BERT in the TF-Hub. BERTOptimizer.py file is modified for freezing the network partially.
Upload ipynb file, BERTOptimizer.py, utils.py files to Google Colab. The data (from Stanford) is downloaded and processed within the code
- Python 3.6
- pandas
- Tensorflow 1.x
- Numpy
- Matplotlib
- Scipy
- Google Colab