Improve note/seg classification

[ninpnin]

Once we have the training, validation and test data (#189), we need to improve the quality of the classification

• [ ] Train and evaluate BERT on the data
• [ ] Run the model on the whole corpus