본문 바로가기
  • Home

Pretrained Model for Civil Complaint Classification

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2026, 31(10), pp.37~45
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : September 14, 2026
  • Accepted : October 13, 2026
  • Published : October 30, 2026

Jongchan Lee 1,  Junha Hwang 1,  Young-Seob Jeong 1

1충북대학교

Accredited

ABSTRACT

With the expansion of online administrative services and the growing volume of civil complaints, automated complaint classification has become increasingly important. However, previous studies have mainly focused on fine-tuning general-purpose language models using labeled data, while the effectiveness of Domain-Adaptive Pretraining (DAPT) using large-scale unlabeled Korean civil complaint and administrative corpora has not been sufficiently investigated. In this study, we constructed a corpus of approximately 137.9 million tokens and applied MLM-based DAPT to A.X-Encoder-base. We also constructed a civil complaint classification dataset consisting of 116,042 queries across 17 categories and compared performance according to the application of DAPT and different training configurations under identical conditions. Experimental results showed that the model trained for 1.25 epochs at a learning rate of 1×10⁻⁴ achieved an accuracy of 73.4% and a Macro F1 score of 72.5%, improving upon the base model by 0.4 and 0.3 percentage points, respectively. These results suggest the importance of domain adaptation in the civil complaint domain.

Journal Copyright Policy

No CCL information provided

Citation status

* References for papers published after 2025 are currently being built.

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