신문사의 정치 성향에 따른 코로나19 보도 내용 분석: 기계학습/딥러닝, 네트워크 분석, 토픽모델링 방법 기반 (Analyzing Partisan News Coverage of COVID-19 with Machine Learning/Deep Learning Algorithms, Network Analysis, and Topic Modeling Methods)

서지정보

· 발행기관 : 한국정보사회학회
· 수록지 정보 : 정보사회와 미디어 / 23권 / 1호 / 69 ~ 105페이지
· 저자명 : 김나현, 이상엽

초록

본 연구의 목적은 신문사의 정치 성향에 따른 코로나19 관련 보도 내용의 차이를 파악하는 것이다. 구체적인 연구 문제는 다음과 같다. 첫 번째로, 언론사의 정치 성향에 따른 보도 내용의 차이가 있는가? 그리고 차이가 존재한다면, 각 언론사들은 코로나19와 관련된 어떠한 이슈를 어느 정도로 다루는가? 두 번째로, 언론사별로 코로나19와 관련하여 사용하는 주요 단어는 어떻게 다른가? 연구 문제에 대한 답을 찾기 위해 본 연구에서는 2020년 2월 12일부터 2021년 1월 31일 사이 조선일보, 동아일보, 경향신문, 한겨레에서 코로나19에 대해서 보도한 기사들을 분석하였다. 분석을 위해 딥러닝 지도학습 알고리즘, 토픽 모델링, 키워드 네트워크에 대한 QAP, 로지스틱 회귀 모형 등을 사용하였다. 분석 결과, 코로나19에 대한 신문사별 보도 내용 차이가 정치 성향에 따라 존재하는 것으로 나타났다. 하지만, 같은 정치 성향을 갖는 신문사들도 다루는 주제들의 비중이 차이가 나는 경우도 있었다. 보수 신문사인 조선일보와 동아일보 경우는 같은 정치 성향이지만 내용의 차이가 존재한 반면 진보 신문사인 경향신문과 한겨레 경우는 내용의 차이가 크지 않은 것으로 나타났다. 본 연구는 다양한 분석 방법 적용을 통해 연구의 간겅성을 제고했으며, 시의적으로 중요한 사회 이슈에 대해 연구했다는 의의를 갖는다. 본 연구에는 코로나19 전체 기간 중 일부 기간에 보도된 기사 내용만을 분석했다는 한계가 존재한다.

영어초록

In the times of crisis, the roles of newspapers are important. In particular, in the case of a new infectious disease like COVID-19, newspapers are known to provide readers with objective information about the disease and important updates that people should know such as new policies imposed by the government. However, newspapers not only play positive roles but also can play negative roles. For example, when they cover more about certain aspects or issues of a disease, it can make readers perceive those aspects or issues more important than others, even though other uncovered or less covered issues are equally or more important. It is known that the news coverage by a news publisher is slanted towards a particular issue depending on the interests of the news publisher. One factor that plays an important role in generating a slant on news coverage is the political orientation of a news publisher. This suggests that news coverage of COVID-19 can be slanted or biased depending on the political orientation of a news publisher, in turn, such slanted coverage can bring about wrong or biased perceptions to readers. this regard, this study examines whether there exist differences in news coverage of COVID-19 according to the political orientation (i.e., conservative or liberal) of a news publisher. We propose two research questions. First, were there differences in news coverage of COVID-19 according to the political orientation of a news publisher? If so, what topics or issues of COVID-19 were more or less covered by a newspaper? Second, what keywords were more or less used by a newspaper? answer these questions, we analyzed news articles about COVID-19 published between February 12, 2020 and January 31, 2021 by the Chosun Ilbo, Dong-A Ilbo, Kyunghyang Shinmun, and Hankyoreh. Using deep learning algorithms, we analyzed whether news coverage of COVID-19 differed according to news publisher's political orientation. For deep learning algorithm, we tried LSTM (Long Short-Term Memory), bidirectional LSTM, and CNN (convolutional neural network). Among these, CNN was finally used as it generated most accurate results. In turn, what topics or issues were more or less covered by each news publisher, we employed a topic modeling method, which is the Latent Dirichlet Allocation (LDA). In addition, by analyzing the keywords networks with the Quadratic Assignment Procedure (QAP) method, we tried to find whether the differences in the keywords network of each news publisher were statistically significant. Finally, by applying a Logistic regression model to the text dataset, we examined what words were more used by each news publisher according to their political orientation. The overall process of data collection and analyses is shown in Figure 1. For the results, we found that there were differences in news coverage depending on the political orientation of a news publisher. The results also showed differences in news coverage existed even between news publishers with similar political orientations. That is, conservative news publishers, the Chosun Ilbo and Dong-A Ilbo, covered different topics. On the other hand, there was little difference in news coverage between liberal news publishers - the Kyunghyang Shinmun and Hankyoreh. According to the results of topic modeling with LDA, liberal news publishers covered more about economy, whereas the Chosun Ilbo covered more about COVID-19 status in foreign countries and the Dong-A Ilbo covered more about prevention of COVID-19. The results of QAP network analysis showed that the difference between the keyword networks of news publishers with different political orientations was statistically significant while that between news publishers with the same political orientation was relatively small and statistically insignificant. Finally, the results of Logistic regression analysis revealed that the liberal news publishers used more of 'laborer', whereas the conservative news publishers used more of 'Wuhan, ' which is a city in China where a novel coronavirus was first discovered. In this study, we examined whether there were differences in news coverage between news publishers with different political orientations with a number of sophisticated computational and statistical analysis approaches. The results of the analyses showed that there were differences, which implies that news publishers should provide objective and unbiased news articles about COVID-19 so that readers can acquire unbiased information about COVID-19 related issues and prevent from spreading the virus more properly.

참고자료

· 없음