Closed brucewlee closed 2 years ago
Also to add, if the labels are given 0 or 1 according to the respective passage's readability, how is NPRM so different from simple pairwise classification?
Hi Bruce,
Re: rank_neural_networks, that is a mistake, I will remove it soon. There is nothing there that affects how any of the code would work.
Re: simple pairwise classification, you are right, NPRM would be an instance of pairwise classification. The novelty of our paper is that we propose using a fine-tuned BERT for pairwise classification; we also construct our labels through a sampled ordering to create the binary labels. We also compared pairwise classification methods using pairwise SVM and various word-embeddings as features.
Thanks for the answer!
Hello. I read your ACL paper. It really is good work. While looking through your code, I found the following line. What is "rank_neural_networks"? I don't find it in PyPI either.
from rank_neural_networks import BertRank, BertRank2