yacis-electra-small-japanese

yacis-electra-small

This is ELECTRA Small model for Japanese pretrained on 354 million sentences / 5.6 billion words of YACIS blog corpus.

The corpus was tokenized for pretraining with MeCab. Subword tokenization was done with WordPiece.

Model architecture

This model uses ELECTRA Small model settings, 12 layers, 128 dimensions of hidden states, and 12 attention heads.

Vocabulary size was set to 32,000 tokens.

Training data and libraries

YACIS-ELECTRA is trained on the whole of YACIS blog corpus, which is a Japanese blog corpus containing 5.6 billion words in 354 million sentences.

The corpus was originally split into sentences using custom rules, and each sentence was tokenized using MeCab. Subword tokenization for pretraining was done with WordPiece.

We used original ELECTRA repository for pretraining. The pretrainig process took 7 days and 6 hours under the following environment: CPU: Intel Core i9-7920X, RAM: 132 GB, GPU: GeForce GTX 1080 Ti x1.

Licenses

The pretrained model with all attached files is licensed under CC BY-SA 4.0, or Creative Commons Attribution-ShareAlike 4.0 International License.

<a rel="license" href="http://creativecommons.org/licenses/by-sa/4.0/"><img alt="Creative Commons License" style="border-width:0" src="https://i.creativecommons.org/l/by-sa/4.0/88x31.png" /></a>

Citations

Please, cite the model using the following citation.

@inproceedings{shibata2022yacis-electra,
  title={日本語大規模ブログコーパスYACISに基づいたELECTRA事前学習済み言語モデルの作成及び性能評価}, 
%  title={Development and performance evaluation of ELECTRA pretrained language model based on YACIS large-scale Japanese blog corpus [in Japanese]}, %% for English citations
  author={柴田 祥伍 and プタシンスキ ミハウ and エロネン ユーソ and ノヴァコフスキ カロル and 桝井 文人}, 
%  author={Shibata, Shogo and Ptaszynski, Michal and Eronen, Juuso and Nowakowski, Karol and Masui, Fumito},  %% for English citations
  booktitle={言語処理学会第28回年次大会(NLP2022) (予定)}, 
%  booktitle={Proceedings of The 28th Annual Meeting of The Association for Natural Language Processing (NLP2022)},  %% for English citations
  pages={1--4},
  year={2022}
}

The model was build using sentences from YACIS corpus, which should be cited using at least one of the following refrences.

@inproceedings{ptaszynski2012yacis,
  title={YACIS: A five-billion-word corpus of Japanese blogs fully annotated with syntactic and affective information},
  author={Ptaszynski, Michal and Dybala, Pawel and Rzepka, Rafal and Araki, Kenji and Momouchi, Yoshio},
  booktitle={Proceedings of the AISB/IACAP world congress},
  pages={40--49},
  year={2012},
  howpublished = "\url{https://github.com/ptaszynski/yacis-corpus}"
}

@article{ptaszynski2014automatically,
  title={Automatically annotating a five-billion-word corpus of Japanese blogs for sentiment and affect analysis},
  author={Ptaszynski, Michal and Rzepka, Rafal and Araki, Kenji and Momouchi, Yoshio},
  journal={Computer Speech \& Language},
  volume={28},
  number={1},
  pages={38--55},
  year={2014},
  publisher={Elsevier},
  howpublished = "\url{https://github.com/ptaszynski/yacis-corpus}"
}