You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Issue:
I am using the ner_ontonotes_bert_mult model to predict entities for text. For sentences with interpunction in the entities, this gives unexpected results. Before the 1.0.0 release, I used the Deeppavlov docker image with the ner_ontonotes_bert_mult config as well. I didn't encounter these issues with the older version of Deeppavlov.
fromdeeppavlovimportbuild_modeldeeppavlov_model=build_model(
"ner_ontonotes_bert_mult",
install=True,
download=True)
sentence='Today at 13:10 we had a meeting'output=deeppavlov_model([sentence])
print(output[0])
[['Today', 'at', '13', ':', '10', 'we', 'had', 'a', 'meeting']]
print(output[1])
[['O', 'O', 'B-TIME', 'O', 'B-TIME', 'O', 'O', 'O', 'O']]
As you can see 13:10 is not recognized as a time entity as a whole, but 13 as B-TIME, : as O, and 10 as B-time. The same happens for names with interpunctions such as E.A. Jones. I also tried the ner_ontonotes_bert configuration, but this gave the same results. Since I want to use the model also for languages other than English, this is not an option at all.
I already opened an issue about this problem. However, the issue was closed without giving me a satisfying outcome.
I was wondering what I could do to solve this issue, is it possible to fine-tune the model on such examples?
The text was updated successfully, but these errors were encountered:
Hello, @ronaldvelzen. Thank you for your interest!
We are sorry that you have encountered the described problem. It turned out that incorrect labeling is related to the specific markup of the training data. We have trained a new model, which is available in pull request, and will be added to the main branch soon.
DeepPavlov version: 1.1.1
Python version: 3.10
Operating system: Ubuntu
Issue:
I am using the ner_ontonotes_bert_mult model to predict entities for text. For sentences with interpunction in the entities, this gives unexpected results. Before the 1.0.0 release, I used the Deeppavlov docker image with the ner_ontonotes_bert_mult config as well. I didn't encounter these issues with the older version of Deeppavlov.
Content or a name of a configuration file:
Command that led to the unexpected results:
As you can see 13:10 is not recognized as a time entity as a whole, but 13 as B-TIME, : as O, and 10 as B-time. The same happens for names with interpunctions such as
E.A. Jones
. I also tried the ner_ontonotes_bert configuration, but this gave the same results. Since I want to use the model also for languages other than English, this is not an option at all.I already opened an issue about this problem. However, the issue was closed without giving me a satisfying outcome.
I was wondering what I could do to solve this issue, is it possible to fine-tune the model on such examples?
The text was updated successfully, but these errors were encountered: