Abstract
In this paper, we suggest a minimally supervised approach for identifying nuanced frames in news article coverage of politically divisive topics. We suggest to break the broad policy frames suggested by Boydstun et al., 2014 into fine-grained subframes which can capture differences in political ideology in a better way. We evaluate the suggested subframes and their embedding, learned using minimal supervision, over three topics, namely, immigration, gun-control, and abortion. We demonstrate the ability of the subframes to capture ideological differences and analyze political discourse in news media.
Highlights
As the political climate and the news media in the United States become increasingly polarized (Prior, 2013; Pew Research Center, 2018), it is important to understand the perspectives underlying the political divisions and analyze their differences
We used the hyper-partisan news dataset (Kiesel et al, 2019) and we crawled additional news articles on the topics from sources with known political bias provided by mediabiasfactcheck.com, where the articles are categorized based on their topics on the websites of the sources
We identify the topic of news articles in the hyper-partisan news dataset by looking at presence of certain keywords in the titles and urls of news articles
Summary
As the political climate and the news media in the United States become increasingly polarized (Prior, 2013; Pew Research Center, 2018), it is important to understand the perspectives underlying the political divisions and analyze their differences. The two articles capture opposite political perspectives, liberal (top) and conservative (bottom) They do not directly contradict each other, instead they focus the discussion on different aspects helping them argue their case. Previous work by Boydstun et al (2014) studied policy issue framing on news media and suggested 15 broad frames to analyze how issues are framed, which include economic, morality and security, among others These framing dimensions can help capture ideological splits (Johnson and Goldwasser, 2016; Johnson et al, 2017b). We use external knowledge sources to identify relevant subframes for each topic and rely on human judgements to match the repeating expressions with these subframes We exploit this resource to train an embedding model, which represents in the same space the subframes labels, the lexicon containing subframes indicator expressions and paragraphs extracted from news articles containing these expressions. We use the model to analyze the different perspectives in left and right leaning news coverage, and their change over time
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have