search

INK-USC / TriggerNER

TriggerNER: Learning with Entity Triggers as Explanations for Named Entity Recognition (ACL 2020)
https://arxiv.org/abs/2004.07493
173 stars 19 forks source link

bug #7

Closed wjczf123 closed 4 years ago

wjczf123 commented 4 years ago

Hi, I think your code may have some problems about data split and model:

  1. In line 101 of semi_supervised.py, 0.2 may be means percentage? I think you should use args.percentage.
  2. In supervised.py, there is no code about data split. Percentage doesn't work at there.
  3. In your paper, you use CNN-BiLSTM model. But your code use BiLSTM as char-level model. I hope you can check it. Thank you a lot for your code.
wjczf123 commented 4 years ago

I use command: python supervised.py --device cuda:0 --dataset CONLL --percentage 3. The result is 85.56. I think the bug should mainly focus on the data (training data is not 3%).

wjczf123 commented 4 years ago

naive.py is normal. supervised.py and semi_supervised.py have problem.

wjczf123 commented 4 years ago

Hope you can fix it as soon as possible.

danny911kr commented 4 years ago

Hi @wjczf123,

Sorry for the late reply.

Thanks for pointing it out. We will add a new argument --unlabeled_percentage for controlling the percentage of the unlabeled data we used for semi-supervised training. We will add a new script to generate the splits and the use of the --percentage argument to load the respective data split for both supervised.py and semi-supervised.py, by the end of this week.

Yeah, it is indeed the BiLSTM model for char-modeling, we will update the description in the paper. (We also tried CNN before, it performed slightly worse. )

Thanks again.

wjczf123 commented 4 years ago

OK. Thank you.

wjczf123 commented 4 years ago

Hope you can fix it as soon as possible.

danny911kr commented 4 years ago

Hi @wjczf123,

I apologize for the late reply due to other project deadlines.

We just updated the repo. Thank you for your suggestion.

Here are the tips for running the code