지도기계학습을 이용한 트위터 뉴스의 프레임 특성 분석 코로나19 보도 프레임의 자동화 판별 방법을 중심으로

초록

Using the supervised machine learning model, this study examined the coverage of COVID-19 by news frames—severity, susceptibility, benefits, and barriers—drawing upon the health belief model. In particular, linguistic features of each frame were automatically derived from the headline and lead of news and they were applied to explore framing features of Twitter messages posted by the major newspapers in South Korea. The data included news articles and tweets about COVID-19 from Chosun Ilbo, JoongAng Ilbo, Kyunghyang Shinmun, and Hankyoreh. To automatically identify news frames, we employed support vector machine(SVM) and naïve Bayes(NB) algorithms by evaluating the accuracy of classifying each frame in 2,000 randomly sampled articles. Furthermore, the optimal classification algorithm was applied to 2,000 randomly sampled tweets to evaluate the predication accuracy for each frame and reveal distinctive linguistic features of each frame on Twitter. Findings showed that perceived threat frames of severity and susceptibility were emphasized in the coverage to a greater extent than behavioral evaluation frames of benefits and barriers, highlighting the risk-aware aspects of Covid-19 prioritized over the costs and benefits of preventive behavior. But we also found that severity and barriers frames were not constructed by consistent and distinct features, given the reduced accuracy of models compared to susceptibility and benefits frames. Furthermore, news frames on Twitter were constructed in a more flexible and discriminatory manner insofar as the logic of social media engages more personalized and emotional use of language in the content. This study sets out a methodology whereby machine learning is employed to code news frames in large-scale news coverage of COVID-19 and identify the features of framing language in an automated, transparent, and reproducible way.

키워드