nlpcl-lab / bert-event-extraction Goto Github PK

Hi, I'm try to reproduce your model. But my result is low. I have checked these labels that my model predicted and I found a lot of labels that was predicted to Event sub-type difference to tag "O" but was tagged to 'O' tag in the dataset. Therefore, my precision score is downgrade( I only get precison=62%) . Did you encountered with this issue. If so, how did your tackled with it. You fixed wrong label in test, dev sets or keep the original data to evaluate these score?
Hope to see your answer soon! Thank you so much!

 1

Originally posted by @mzh1996 in #15 (comment)

阅读代码发现对arguments的预测依赖于grountruth，打印预测的arguments发现start_idx和end_idx都与groundtruth相同。请问是我使用的方式不对吗？

Hi,
I read your code and found there are two problems that hinder the performance improvement.
First, as I know, previous papers use head words of entity mentions as the candidate arguments, but you use the whole word sequence of entity mentions, which harms the argument-level performance a lot.
Second, while training, you train the argument-level classifier based on predicted triggers, instead, I believe the argument-level classifier should be trained on the golden triggers.

Thanks for your sharing！ But where is eval.py？

TypeError: new() received an invalid combination of arguments - got (NoneType, int), but expected one of:

• (*, torch.device device)
  didn't match because some of the arguments have invalid types: (!NoneType!, !int!)
• (torch.Storage storage)
• (Tensor other)
• (tuple of ints size, *, torch.device device)
• (object data, *, torch.device device)

hi @bowbowbow, I am doing the end-of-the-year check up.
as i see it, i recognize that you tried some new approaches
to extract events while using bert.
I believe that the new approach can be captured by the NN architecture diagram

if it is possible, please add the diagram so that people can catch the new approach in a glance

thanks, yours sincerely, Yang

Hi,
Thank you for sharing.
I'm interested if you tried to use bert to improve the performance of JMEE.
I try to reproduce JMEE,but I can't achieve the result of paper.

Traceback (most recent call last):
File "D:/pythonProject/bert-event-extraction-master/train.py", line 80, in
model = Net(
File "D:\pythonProject\bert-event-extraction-master\model.py", line 18, in init
self.entity_embed = MultiLabelEmbeddingLayer(num_embeddings=entity_size, embedding_dim=entity_embedding_dim, device=device)
File "D:\pythonProject\bert-event-extraction-master\model.py", line 129, in init
self.matrix = nn.Embedding(num_embeddings=num_embeddings,
File "D:\anaconda2020\lib\site-packages\torch\nn\modules\sparse.py", line 109, in init
self.weight = Parameter(torch.Tensor(num_embeddings, embedding_dim))
TypeError: new() received an invalid combination of arguments - got (NoneType, int), but expected one of:

• (*, torch.device device)
  didn't match because some of the arguments have invalid types: (!NoneType!, !int!)
• (torch.Storage storage)
• (Tensor other)
• (tuple of ints size, *, torch.device device)
• (object data, *, torch.device device)

pip install pytorch==1.0 pytorch_pretrained_bert==0.6.1 numpy
下载一直失败，我人都麻了，能不能来个大佬帮帮我

您好，请问如何获取ACE2005数据集呢？需要什么lincense吗？有了这个lincense之后可以去哪里下载或者申请呢？多谢

x是[batch_size，SEQ_LEN，768]的bert表达
有一句代码：
for i in range(batch_size):
x[i] = torch.index_select(x[i], 0, head_indexes_2d[i])
请问这是在做什么？

不知道您读过这篇文章没有：《Exploring Pre-trained Language Models for Event Extraction and Generation
Sen》

他直接吧trigger的识别准确率推到了80%

我跑了50轮，最高trigger classification只有23的f1。不知道67分作者使用的是什么参数

I'm a freshman in Event Extraction. I have learned your code. In the train.py, I think this is a multitask because the loss is the sum of triggers loss and arguments loss. So i don't is this model a joint method or pipeline method?

hi @bowbowbow, I am doing the end-of-the-year check up.
it processes the preprocessed data, but some people may utilize it for a sentence
that they want to process.

is it possible to input a sentence and check the output events within the input sentence?
if so, please add the description for it in README

thanks, yours sincerely, Yang