Classify with multiple sentences

[siccokoetsier]

Hi,

I am pretty much a beginner in NLP and I am trying to build a basic binary BERT-classifier based on new Dutch text data + a tagged sentiment value in the training set. Yet in my case, each text item consists of multiple sentences. Most online examples only treat one-lined cases. I have the impression that Bert is very sensitive to this and needs to be fed in a very specific way.

How can I train the model with multiple sentence items? I do find some reference to this problem in the context of RepresentationModel ('encode_sentences'), but it's very hard for me to apply and integrate this code into the general classifier examples with ClassificationModel (f.e. here and here ).

Thanks for any hints! Feel free to contact me anytime. Keep up the good work with this excellent lib!

[luizniero]

Hello. I have suggest a code for this. You can find my code at: #1360