Roberta Large Ner English

🚀 roberta-large-ner-english: model fine-tuned from roberta-large for NER task

This is an English NER model fine-tuned from roberta-large on the conll2003 dataset. It performs well on emails/chat data and is especially effective for entities not starting with an uppercase letter.

🚀 Quick Start

This section will guide you on how to use roberta-large-ner-english with HuggingFace.

Load the model and tokenizer

from transformers import AutoTokenizer, AutoModelForTokenClassification

tokenizer = AutoTokenizer.from_pretrained("Jean-Baptiste/roberta-large-ner-english")
model = AutoModelForTokenClassification.from_pretrained("Jean-Baptiste/roberta-large-ner-english")

Process text sample

from transformers import pipeline

nlp = pipeline('ner', model=model, tokenizer=tokenizer, aggregation_strategy="simple")
nlp("Apple was founded in 1976 by Steve Jobs, Steve Wozniak and Ronald Wayne to develop and sell Wozniak's Apple I personal computer")

[{'entity_group': 'ORG',
  'score': 0.99381506,
  'word': ' Apple',
  'start': 0,
  'end': 5},
 {'entity_group': 'PER',
  'score': 0.99970853,
  'word': ' Steve Jobs',
  'start': 29,
  'end': 39},
 {'entity_group': 'PER',
  'score': 0.99981767,
  'word': ' Steve Wozniak',
  'start': 41,
  'end': 54},
 {'entity_group': 'PER',
  'score': 0.99956465,
  'word': ' Ronald Wayne',
  'start': 59,
  'end': 71},
 {'entity_group': 'PER',
  'score': 0.9997918,
  'word': ' Wozniak',
  'start': 92,
  'end': 99},
 {'entity_group': 'MISC',
  'score': 0.99956393,
  'word': ' Apple I',
  'start': 102,
  'end': 109}]

✨ Features

• Fine-tuned on conll2003: [roberta-large-ner-english] is fine-tuned from roberta-large on the conll2003 dataset.
• Good performance on specific data: Validated on emails/chat data, it outperforms other models on this type of data, especially for entities not starting with an uppercase letter.

📦 Training data

Training data was classified as follows:

To simplify, the prefix B- or I- from the original conll2003 was removed. The train and test datasets from the original conll2003 were used for training, and the "validation" dataset was used for validation, resulting in a dataset of the following sizes:

📚 Documentation

Model performances

On conll2003 validation dataset (computed on the tokens predictions)

On private dataset (email, chat, informal discussion) (computed on word predictions)

Comparison with Spacy (en_core_web_trf - 3.2.0) on the same private dataset

Related article

For those interested, here is a short article on how the results of this model were used to train an LSTM model for signature detection in emails: LSTM model for email signature detection

📄 License

This project is licensed under the MIT license.