Comparison of KoBERT and BERT for Emotion Classification of Healthcare Text Data

In recent times, the rapid progress of digital technology has led to a substantial increase in the popularity of digital health. Identifying depression, which is a prevalent mental illness, is crucial in digital healthcare to prevent further harm and provide timely support. This study proposes an AI model that automates the identification of depressive patients. By leveraging Natural Language Processing (NLP) and pre-trained language models like BERT, we aim to classify emotions into six categories. Training the model requires a Korean emotional conversation corpus, which we obtain through crowd-sourcing and AI-Hub’s user case studies. To extend the applicability to English-speaking countries, we plan to translate the Korean corpus using the Google Translation API and fine-tune the BERT model with English data. The feasibility of the English model was evaluated by comparing the performance of KoBERT and BERT in emotion understanding. The findings will offer valuable insights into these models’ efficacy and contribute to the field of emotion classification.