Toward Better Sentence Embedding with BERT

Problem When it comes to comparing two sentences, the original BERT model suggests concatenating two sentences together separated by [SEP] token, and feeding that long sentence into the model to get the score. But as you may have guessed, that’s not ...