CLoSE: Contrastive Learning of Subframe Embeddings for Political Bias Classification of News Media

Proceedings of the 29th International Conference on Computational Linguistics, pages 2780–2793 October 12–17, 2022. 2780 CLoSE: Contrastive Learning of Subframe Embeddings for Political Bias Classification of News Media Michelle YoungJin Kim and Kristen Marie Johnson Computer Science and Engineering, Michigan State University, USA {kimmic16,kristenj}@msu.edu Abstract Framing is a political strategy in which journal- ists and politicians emphasize certain aspects of a societal issue in order to influence and sway public opinion. Frameworks for detect- ing framing in news articles or social media posts are critical in understanding the spread of biased information in our society. In this paper, we propose CLoSE, a multi-task BERT- based model which uses contrastive learning to embed indicators of frames from news articles in order to predict political bias. We evaluate the performance of our proposed model on sub- frames and political bias classification tasks. We also demonstrate the model’s classification accuracy on zero-shot and few-shot learning tasks, providing a promising avenue for fram- ing detection in unlabeled data. 1 Introduction News media coverage shapes our attitudes, emo- tions, and decisions toward public issues (Iyen- gar, 1994; Pan and Kosicki, 1993; Jensen et al., 2014). Research shows that people’s perceptions of news can be manipulated by presenting the same story with different expressions. For exam- ple, participants of a study found a terrorist attack caused by “al-Qaeda and associated radical Islamic groups" considerably more threatening than a ter- rorist attack by “domestic rebel separatist groups,” which is an equivalent paraphrase (Kapu´sci´nski and Richards, 2016). Hence, studies on the influences of different presentations of issues, or the effects of framing, play an essential role in understanding political discourse. Framing refers to emphasizing desired aspects of an issue to promote and amplify a particular per- spective (Entman, 1993). By selecting and thus elevating the salience of a specific angle of a topic, media sources can present the topic through their choice of frames to induce particular attributes and judgments among the public. Framing is widely re- searched on various topics, from its effects on pub- lic opinion on political issues such as the U.S. anti- nuclear war movement (Entman and Rojecki, 1993) and stem cell research (Nisbet et al., 2003) to the economic impact of framing (Liu and Pennington- Gray, 2015; Van Dalen et al., 2017). In this work, we study framing detection for three politically polarized issues in the U.S. news media: abortion, gun control, and immigration. We focus on framing in the news discourse to under- stand a discrepancy in media consumption and its influence on political bias polarization. With the increase of news media outlets and social media platforms, the public is overwhelmed with a flood of information. Unfortunately, the increase of news sources does not yield the sharing of information across political views, often developing into biased echo chambers on social media platforms. In fact, there are stark differences in the social media and news sources that liberals and conservatives use and trust. Fox News is the primary news source for nearly half …

Content truncated to 500 words