본문 바로가기
  • Home

Optimizing Emotion Recognition in Korean Counseling Texts Using Parameter Efficient Fine Tuning

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2026, 31(5), pp.75~84
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : April 29, 2026
  • Accepted : May 19, 2026
  • Published : May 29, 2026

Myung-Suk Lee 1

1계명대학교

Accredited

ABSTRACT

This study proposes a parameter-efficient emotion recognition model optimized for Korean counseling dialogues to address the demand for online psychological crisis response. Parameter-efficient fine-tuning (PEFT) techniques, specifically LoRA and QLoRA, were applied to KcBERT and evaluated against full fine-tuning and KoGPT2 using the AI Hub(Artificial Intelligence Hub) corpus. Results show that LoRA-based KcBERT achieved 0.824 accuracy and 0.812 F1-score while updating only 0.3% of parameters. t-SNE analysis confirmed stable emotional representation learning. This research demonstrates that high-performance emotion recognition can be practically deployed in resource-constrained environments, significantly improving the efficiency of AI-based mental health support.

Citation status

* References for papers published after 2024 are currently being built.

This paper was written with support from the National Research Foundation of Korea.