sayarghoshroy/summaformers

Code for our Paper, 'Summaformers @ LaySumm 20, LongSumm 20' at EMNLP 2020, Scholarly Document Processing Workshop

summaformers's Issues

Some question about the model

Dear author, your work is outstanding! I read your paper carefully, and I would like to consult some details about some questions.
In the LongSumm task, how does the budget module of your model calculate the weights of summary sentences obtained in different sections?
Here is the statement extracted from your paper:
" The best performing setting corresponds to selecting sections whose ROUGE-1 overlap with the long summary
is greater than 20.0. Intuitively, this prunes out irrelevant sections such as ‘abbreviations’ and ‘acknowledgements’. The remaining sections were assigned weights based on the ROUGE-1 overlap with the provided long summary. "

I don't understand that meaning. When you test your model and there is no Longsum as target, how do you compute the rouge-1 between different sections with the target?