연구개발정보 자동분류를 위한 분류모델 비교 연구

A Comparative Study of Classification Models for Automatic Classification of Research and Development Information
  • 김원영; 
  • 최예지; 
  • 이혜성; 
  • 이서진; 
  • 장항배

초록

Research information produced through national R&D projects continues to accumulate. However, much of it is embedded in free-form documents such as R&D reports, which limits its systematic use. This study constructed a sentence-level dataset from 41 R&D reports obtained through NTIS. It also developed a model comparison framework for auto matically classifying R&D information into 11 information types. GPT-5.5-based labeling was applied, and the training data were refined using a confidence threshold. The study then compared a TF-IDF+LinearSVC baseline with Korean pretrained language model-based classifiers. The results show that the Korean pretrained language model-based approach outperformed the traditional machine learning approach. In particular, the klue/roberta-base model achieved the best performance. These findings confirm the feasibility of sentence-level structuring of free-form research information in R&D reports. They also provide a basis for advancing future R&D information rating and utilization systems.

키워드

연구개발정보; 정보 등급화; 문서 보안; 사전학습 언어모델; R&D Information; Information Classification; Document Security; Pre-trained Language Model
제목
연구개발정보 자동분류를 위한 분류모델 비교 연구
제목 (타언어)
A Comparative Study of Classification Models for Automatic Classification of Research and Development Information
저자
김원영; 최예지; 이혜성; 이서진; 장항배
DOI
10.7838/kjdi.2026.31.3.025
발행일
2026-08
유형
Y
저널명
한국디지털산업학회지
권
31
호
3
페이지
25 ~ 40