Modeling Political Framing Across Policy Issues and Contexts

Predicting Political Frames Across Policy Issues and Contexts Shima Khanehzar Andrew Turpin Gosia Mikolajczak School of Computing and Information Systems The University of Melbourne skhanehzar@student.unimelb.edu.au {aturpin,malgorzata.mikolajczak}@unimelb.edu.au Abstract Politically-contested issues are often discussed with different emphases by different people. This emphasis is called a frame. In this pa- per, we examine the performance of classifiers trained using the media frames Corpus (MFC) (Card et al., 2015); a collection of US news la- belled with fifteen different frame categories. Specifically, we compare pre-trained language models (XLNet, Bert, and Roberta), fine-tuned using MFC, against results from the literature and simpler models in their ability to predict frames from text. We also test these models on a new corpus that we have derived from Australian parliamentary speeches. Our exper- imental results first show that the fine-tuned models significantly outperform the current best methods on MFC. We also show that the model fine-tuned on US news articles can be convincingly applied to predict policy frames in Australian parliamentary speeches, though the accuracy is significantly reduced, suggest- ing potential discrepancy in framing strategies and/or text usage between US News and Aus- tralian Parliamentary Speeches. 1 Introduction Politicians and the media often portray political issues in a subjective way in an attempt to shape public attitudes (Chong and Druckman, 2007). For example, a politician opposing the same-sex mar- riage (SSM) might frame the issue using the lens of tradition and religious beliefs, whereas a politi- cian supporting SSM might frame a speech us- ing fairness and equality as the base. Due to its complexity and linguistic subtleties, issue fram- ing (Entman, 1993) remains challenging for au- tomated text methods. To address these chal- lenges, recent work by Boydstun et al. (2013) de- fines broad categories of common policy frames and annotates US News articles to build the media frames Corpus (MFC) (Card et al., 2015). Follow- up studies have used the MFC to investigate the ac- curacy of models that attempt to classify the dom- inant frames of US news articles. In this paper, we aim to extend this work and answer the following question: can recent pre-trained neural classifiers learn to predict dominant frames across issues and communication contexts? To answer this question, we provide the following contributions. • We investigate the effectiveness of the pre- trained language models XLNet, Bert and Roberta in predicting dominant frames within each issue on the MFC. • We investigate whether our models can learn to predict frame categories across issues. Our results show that we can apply trained models on a new issue without training data for that particular issue. • We annotate a small subset of Australian par- liamentary speeches on Same-Sex Marriage (SSM). • We evaluate whether our models can learn to predict frames across communication con- texts, applying the models fine-tuned on the MFC dataset on the Australian parliamentary speeches. 2 Background and Related Work Natural Language Processing techniques have been applied to identify several aspects of the po- litical discourse including ideology (Iyyer et al., …

Content truncated to 500 words