심층 주제, 지역, 장르를 모두 분류할 수 있는 다면적 뉴스 기사 자동 분류 모델 연구

Abstract

This study developed a model to classify news articles into categories of topic, genre, and region using a Korean Pre-trained Language model. To achieve this, a new news article classification system was designed by referring to the classification systems of domestic media outlets. The topic and genre classification models were implemented as hierarchical classification models that link the main categories and subcategories, and their performance was compared with that of an integrated category model. The evaluation results showed that the hierarchical structure classification model had the advantage of providing more precise categorization in ambiguous or overlapping categories compared to the integrated category model. For regional classification of news articles, a model was built to classify into 18 categories, and for regional news articles, the regional characteristics were clearly reflected in the text, resulting in high performance. This study demonstrated the effectiveness of classifying news articles from multiple perspectives—topic, genre, and region—and emphasized the significance of suggesting the potential for a multi-dimensional news article classification service that meets user needs.