A Media Frame Analysis of Global Warming Articles Using Natural Language Processing

Overview

Global warming (GW) has recently been viewed as one of the most pressing environmental issues and has been hotly discussed in politics. Growing concerns and controversy over GW has attracted increasing media attention towards this issue. As the media is a major influencer of public opinion, it is important to understand how the media presents or frames GW. This study sought to develop machine learning models for frame detection, identify the best-performing model, and use that model to study how GW is presented in a dataset of ~56K articles. After training a naïve bayes, logistic regression, and BERT model on annotated data, the logistic regression model achieved the highest accuracy (63%) on previously seen issues, while the BERT model achieved the highest accuracy (47%) previously unseen issues. All model performances exceeded the baseline accuracy of 6.67% expected by random classification. Applying the BERT model to the GW dataset revealed temporal framing fluctuations reflective of major events like presidential elections. Further, articles from liberal and conservative outlets exhibited different framing trends, yet the two shared similar dominant frames (i.e. economic, cultural identity, and health and safety). These results can help news sources decide how to cover GW and aid politicians in determining policy agendas. This study also highlights the superiority of deep learning models over traditional algorithms when applied to previously unseen issues, while questioning whether such advantages of deep learning are cost-worthy.

Competition history

• JSHS 2022 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.