Macbert Base Chinese Medical Collation

🚀 MacBert-based Medical Spell-checking Model

This model is a fine - tuned version of MacBert for spell - checking in medical application scenarios, achieving high accuracy on a large medical dataset.

🚀 Quick Start

You can use this model directly with a pipeline for token classification:

>>> from transformers import (AutoModelForTokenClassification, AutoTokenizer)
>>> from transformers import pipeline

>>> hub_model_id = "9pinus/macbert-base-chinese-medical-collation"

>>> model = AutoModelForTokenClassification.from_pretrained(hub_model_id)
>>> tokenizer = AutoTokenizer.from_pretrained(hub_model_id)
>>> classifier = pipeline('ner', model=model, tokenizer=tokenizer)
>>> result = classifier("如果病情较重，可适当口服甲肖唑片、环酯红霉素片等药物进行抗感染镇痛。")

>>> for item in result:
>>>     if item['entity'] == 1:
>>>         print(item)

{'entity': 1, 'score': 0.58127016, 'index': 14, 'word': '肖', 'start': 13, 'end': 14}

✨ Features

• This is a fine - tuned version of MacBert for medical spell - checking.
• Fine - tuned on a 300M dataset with 60K+ authorized medical articles.
• Randomly confuses 30% of sentences in the articles by adding noise with visually or phonologically resembled characters.
• Achieves 96% accuracy on the test dataset.

📚 Documentation

Model description

This model is a fine - tuned version of MacBert for the purpose of spell checking in medical application scenarios. We fine - tuned MacBert Chinese base version on a 300M dataset including 60K+ authorized medical articles. We proposed to randomly confuse 30% sentences of these articles by adding noise with a either visually or phonologically resembled characters. Consequently, the fine - tuned model can achieve 96% accuracy on our test dataset.

Intended uses & limitations

You can use this model directly with a pipeline for token classification as shown in the quick start section.

Framework versions

• Transformers 4.15.0
• Pytorch 1.10.1+cu113
• Datasets 1.17.0
• Tokenizers 0.10.3

📄 License

This model is licensed under the Apache 2.0 license.

📦 Additional Information