Practical usage of KcBERT-nsmc

• Korean sentiment classifier based on pre-trained KcBERT and NSMC data (for fine tuning)
• Accuracy is 0.885 with decreased learning rate and about 2500 steps of fine tuning
• Load fine-tuned model on Flask for user interface - Provides a softmax score and argmax score for input sentence.

References

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