Model Specification

• This is the state-of-the-art Twitter NER model (with 74.35% Entity-Level F1) on Tweebank V2's NER benchmark (also called Tweebank-NER), trained on the corpus combining both Tweebank-NER and WNUT 17 training data.
• For more details about the TweebankNLP project, please refer to this our paper and github page.
• In the paper, it is referred as HuggingFace-BERTweet (TB2+W17).

How to use the model

• PRE-PROCESSING: when you apply the model on tweets, please make sure that tweets are preprocessed by the TweetTokenizer to get the best performance.

from transformers import AutoTokenizer, AutoModelForTokenClassification

tokenizer = AutoTokenizer.from_pretrained("TweebankNLP/bertweet-tb2_wnut17-ner")

model = AutoModelForTokenClassification.from_pretrained("TweebankNLP/bertweet-tb2_wnut17-ner")

References

If you use this repository in your research, please kindly cite our paper:

@article{jiang2022tweetnlp,
    title={Annotating the Tweebank Corpus on Named Entity Recognition and Building NLP Models for Social Media Analysis},
    author={Jiang, Hang and Hua, Yining and Beeferman, Doug and Roy, Deb},
    journal={In Proceedings of the 13th Language Resources and Evaluation Conference (LREC)},
    year={2022}
}