악성 댓글 기반 이슈 분류를 위한 토픽 클러스터링 연구
A Study on Topic Clustering for Issue Classification Based on Malicious Comments
- 발행
- 2026 등록KCI API에는 발행일 항목이 없어 원문 등록일을 쓰고 있습니다. 실측으로 발행보다 최대 6년 늦습니다 — 재수집하면 발행연월로 바뀝니다.
- 소속·발행
- 국립한밭대학교 소프트웨어융합대학원 정보통신학과
- 출처
- 국내 KCI
- 원문
- 원문 보기 ↗
개념
키워드
Malicious Comments, BERTopic, Topic Clustering, Co-occurrence Network, Time Series, Malicious Comments, BERTopic, Topic Clustering, Co-occurrence Network, Time Series
초록
Purpose This study aims to analyze online malicious comments not as simple emotional expressions but as issue-based discourse structures, quantitatively identifying the formation and diffusion patterns of social conflicts.
 Design/Methodology/Approach A total of 59,654 comments from the KBS News YouTube channel were collected between May and September 2025. A dictionary of 113 malicious keywords was constructed, and ten major topics were derived using Sentence-BERT-based BERTopic modeling. The core keywords of each topic were visualized through co-occurrence network (SNA) analysis, and weekly weight variations were examined using a heatmap-based time series analysis.
 Findings "Political conflict and power criticism" and "media distrust" emerged as the dominant issues, while keywords such as rebellion, impeachment, fake news, and fraud functioned as central hubs in the network. Particularly during the third week of June and the second week of September, an issue-reactive pattern appeared, indicating that malicious comments serve as a structural medium for the spread of public opinion online.