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