Repository files navigation

They are designed to be adapted to the Japanese medical domain.
The medical corpora were scraped for academic use fromToday's diagnosis and treatment: premium, which consists of 15 digital references for clinicians in Japanese published byIGAKU-SHOIN Ltd..
The general corpora were extracted from a Wikipedia dump file (jawiki-20190901) onhttps://dumps.wikimedia.org/jawiki/.

• medBERTjp - MeCab-IPAdic
  • pre-trained model followingMeCab-IPAdic-tokenized Japanese BERT model.
  • Japanese tokenizer:MeCab + Byte Pair Encoding (BPE)
  • ipadic-py, or manual install of IPAdic is required.
  • max_seq_length=128
• medBERTjp - Unidic-2.3.0
  • Japanese tokenizer:MeCab + BPE
  • Unidic v2.3.0+2020-10-08 viaunidic-py is required.
  • max_seq_length=128
• medBERTjp - MeCab-IPAdic-NEologd-JMeDic
  • Japanese tokenizer:MeCab + BPE
  • install of bothmecab-ipadic-NEologd andJ-MeDic (MANBYO_201907_Dic-utf8.dic) is required.
  • max_seq_length=128
• medBERTjp - SentencePiece
  (Old:v0.1-sp)
  • Japanese tokenizer:SentencePiece followingSentencepiece Japanese BERT model
  • use customized tokenization for the medical domain by SentencePiece
  • max_seq_length=128

For just using the models:

• Transformers (>=2.11.0)
• fugashi, a Cython wrapper forMeCab
  • ipadic,unidic-py,mecab-ipadic-NEologd, andJ-MeDic: if required.
• SentencePiece would be automatically installed withTransformers.

Please check code examples oftokenization_example.ipynb, or try to useexample_google_colab.ipynb onGoogle Colab.

This work was supported by Council for Science, Technology and Innovation (CSTI), cross-ministerial Strategic Innovation Promotion Program (SIP), "Innovative AI Hospital System" (Funding Agency: National Institute of Biomedical Innovation, Health and Nutrition (NIBIOHN)).

The pretrained models are distributed under aCreative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0).
They are freely available for academic purpose or individual research, but restricted for commecial use.

The codes in this repository are licensed under the Apache License, Version2.0.