Online Newspaper Framing of Non-Communicable Diseases: Comparison of Mainland China, Taiwan, Hong Kong and Macao

Abstract

As non-communicable diseases (NCDs) are now well recognized as the leading cause of mortality among adult populations worldwide, they are also increasingly the focus of media coverage. As such, the objective of this study is to describe the framing of NCDs in the coverage of newspapers, with the understanding that it says something about the society producing it. Automatic content analysis was employed to examine disease topics, risks, and cost consequences, thus providing lay people with a chance of learning the etiology of NCDs and information available for fighting diseases. The result of the computational method identified a total of 152,810 news articles with one of the seven supra-categories of NCDs. The category of metabolic diseases was covered most frequently in the past ten years. Three health risks received ample attention in all 11 newspapers: stress burden, tobacco use, and genetic predispositions. The results evidenced how media framed risk information of illnesses would distort the way in which diseases were selected, interpreted, and the outcome communicated. Future research building on our findings can further examine whether news framing affects the way the readers perceive and prevent NCDs.

Communicative Blame in Online Communication of the COVID-19 Pandemic: Computational Approach of Stigmatizing Cues and Negative Sentiment Gauged With Automated Analytic Techniques

“…Taiwan’s news media has been described as professional and independent [ 42 ]; however, online news services have been facilitating a negative tone. Online news has persistently used stigmatizing terms to initiate the adoption of stigma-related emotion words.…”

“…It also ensured that the time periods are comparable for operational reasons such as why certain weeks may have higher demand than others or other factors that could influence blaming discourse. The research strategy included enabling DivoMiner to identify data in their natural settings by retrieving and filtering online media sources [ 41 , 42 ] and constructing vocabularies and learning word representations from NLP for analysis along with the research theme (stigmatizing and sentiment language) from online media genres [ 41 - 43 ].…”

“…To ensure the efficiency of capturing opinions related to COVID-19 from unstructured text in syntactically explicit language, data were trained to include some knowledge of semantic meaning in our model [ 42 , 43 , 45 ]. To verify the feasibility and reliability of the word embedding, three coders manually labeled and checked 1500 random postings.…”

“…To verify the feasibility and reliability of the word embedding, three coders manually labeled and checked 1500 random postings. After testing and training, the classification accuracy level reached 75% (1125/1500), which was deemed to be acceptable [ 41 , 42 ]. After irrelevant opinionated data were excluded (eg, shopping, nonnews), the DivoMiner classifier implemented the filtering process of the recorded data in digital form.…”

“…The interrater reliability employed the Cohen kappa coefficient by computing sentiment and targeted figures, groups, and organizations for the online postings. The interrater reliability was 0.5901 (95% CI 0.49-0.71; P =.06) and rated as moderate [ 42 , 43 ].…”