KoBERT-nsmc

• KcBERT를 이용한 네이버 영화 리뷰 감정 분석 (sentiment classification)
• 🤗Huggingface Tranformers🤗 라이브러리를 이용하여 구현

Dependencies

• torch==1.5.1
• transformers~=3.0.1

How to use KcBERT on Huggingface Transformers Library

from transformers import AutoModel, AutoTokenizer

model = AutoModel.from_pretrained('beomi/kcbert-base')
tokenizer = AutoTokenizer.from_pretrained('beomi/kcbert-base')

Usage

$ python3 main.py \
    --model_name_or_path beomi/kcbert-base \
    --do_train --do_eval \
    --max_seq_len 100

Prediction

$ python3 predict.py --input_file {INPUT_FILE_PATH} --output_file {OUTPUT_FILE_PATH} --model_dir {SAVED_CKPT_PATH}

Results

	Accuracy (%)
KcBERT	89.04
KoBERT	89.63
DistilKoBERT	88.41
Bert-Multilingual	87.07
FastText	85.50

References

• Monologg KoBERT for Transformers
• KoBERT
• Huggingface Transformers
• NSMC dataset