search

nlpyang / BertSum

Code for paper Fine-tune BERT for Extractive Summarization
Apache License 2.0
1.47k stars 424 forks source link

About the highlights in CNN/DM datasets #85

Open TTWen opened 4 years ago

TTWen commented 4 years ago

If I understand correctly, according to the paper, you use a greedy selection to choose the oracle summaries (according to the rouge score), and use the oracle summaries in the following training, I want to know if the rouge score in the selection of oracle summary is calculated between highlights and every sentence？and if the process of choosing oracle summary according to rouge score is the only place that uses the highlights data, I mean, the highlights from the CNN/DM datasets will not join the following training？

RafaelWO commented 4 years ago

I can only say that I understood it the same way. So yes :+1: in my opinion.

binhna commented 4 years ago

But when looking at the code, I saw him use the Bertdecoder with these highlights when it comes to the abstractive method