Named entity recognition (NER) is task that mark tags of the input text sequence.
BERT-CRF is a good NER model.
I want to find a better NER model. Or I want to improve the BERT-CRF model.
What can I do?
Hi Tong Guo , you can also try the following:
1.Deep Pavlov
http://docs.deeppavlov.ai/en/master/features/models/ner.html
2. Polyglot https://polyglot.readthedocs.io/en/latest/NamedEntityRecognition.html
3. Flair ( provides ensemble modeling feature too)
https://github.com/flairNLP/flair
I would also encourage you to try multiple models as it depends from task to task.
One model might have faster inference
other may give near-perfect result for particular tags such as name or location.
So best model would be that does your task done more efficiently and may have some tradeoffs.
This one is a good paper I found on the subject https://arxiv.org/pdf/1812.09449.pdf
The way I used to approach this was:
1. To read a survey paper that covers the latest SOTA models for NER.
2. Then Finding of their pros and cons i.e. a comparative analysis.
3. Try them on sample data ( 10-20%) trying to cover most of the cases. ( This was done to confirm the pros and cons mentioned)
4. Figuring out which ones to proceed with.
I would highly encourage you to do the comparative study, you can start by taking any 2 models with your desired one.
- Badges
- Science topic