Few-Shot Named Entity Recognition work

huggingface / setfit

Efficient few-shot learning with Sentence Transformers

https://hf.co/docs/setfit

Apache License 2.0

2.14k stars 217 forks source link

Few-Shot Named Entity Recognition work #72

Open zhanghaok opened 1 year ago

zhanghaok commented 1 year ago

Hi, really like your work, have you considered using this framework for few-shot named entity recognition work? or do you have an example code for it, looking forward to the progress in few-shot named entity recognition!

lewtun commented 1 year ago

Hey @zhanghaok thanks for your interest! We haven't quite figured yet out how to apply SetFit in token classification tasks, but we will certainly add support to the library if we can make it work well :)

zhanghaok commented 1 year ago

Hey @zhanghaok thanks for your interest! We haven't quite figured yet out how to apply SetFit in token classification tasks, but we will certainly add support to the library if we can make it work well :)

Thanks for your reply, look forward to it being used in few-shot named entity recognition , come on!💪

hooman-bayer commented 1 year ago

Hi all, I think it will be a great success to extend this to token-classification too. Have you thought about something along the lines of the followings:

Keybert (https://github.com/MaartenGr/KeyBERT)
TARS (https://github.com/flairNLP/flair/blob/master/resources/docs/TUTORIAL_10_TRAINING_ZERO_SHOT_MODEL.md) which is from FLAIR - Please see (Use Case 2: Zero-shot Named Entity Recognition (NER) with TARS - in the notebook)

I think specially something along the lines of KeyBert or TARS might already be something that is usable

@hooman650