Sentence-level Media Bias Analysis Informed by Discourse Structures

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 10040 - 10050 December 7-11, 2022 ©2022 Association for Computational Linguistics Sentence-level Media Bias Analysis Informed by Discourse Structures Yuanyuan Lei1 Ruihong Huang1 Lu Wang2 Nick Beauchamp3 1Computer Science and Engineering, Texas A&M University, College Station, TX 2Computer Science and Engineering, University of Michigan 3Political Science, Northeastern University 1{yuanyuan, huangrh}@tamu.edu 2wangluxy@umich.edu, 3n.beauchamp@northeastern.edu Abstract As polarization continues to rise among both the public and the news media, increasing at- tention has been devoted to detecting media bias. Most recent work in the NLP community, however, identify bias at the level of individual articles. However, each article itself comprises multiple sentences, which vary in their ideolog- ical bias. In this paper, we aim to identify sen- tences within an article that can illuminate and explain the overall bias of the entire article. We show that understanding the discourse role of a sentence in telling a news story, as well as its relation with nearby sentences, can reveal the ideological leanings of an author even when the sentence itself appears merely neutral. In par- ticular, we consider using a functional news dis- course structure and PDTB discourse relations to inform bias sentence identification, and dis- till the auxiliary knowledge from the two types of discourse structure into our bias sentence identification system. Experimental results on benchmark datasets show that incorporating both the global functional discourse structure and local rhetorical discourse relations can ef- fectively increase the recall of bias sentence identification by 8.27% - 8.62%, as well as in- crease the precision by 2.82% - 3.48%1. 1 Introduction News media play a vast role not only by providing information, but also by selecting, packaging, and organizing the information to shape public opinions (Mccombs and Reynolds, 2002, 2009). Multiple studies showed that the media outlets are becoming more partisan and polarized, with great potential to influence public’s political stance (Gentzkow and Shapiro, 2010b,a), which presents the necessity to develop novel models to detect the media bias. Most recent research work focus on detecting the media bias either at the level of media outlet 1The code link: https://github.com/yuanyuanlei-nlp/ bias_sentence_discourse_emnlp_2022 (Baly et al., 2018), or at the level of individual ar- ticles (Kiesel et al., 2019; Baly et al., 2020a; Roy and Goldwasser, 2020). However, each article it- self comprises multiple sentences, which vary in their ideological bias (Entman, 2006, 2007a). In this paper, we focus on sentence-level media bias analysis to identify bias sentences that, as inter- preted by (Gentzkow and Shapiro, 2006; Entman, 2007b; Mullainathan and Shleifer, 2002), provide the supportive or background information to shift opinion in an ideological direction, though that may be done via selective inclusion or omission as well as overt ideological language. The identified bias sentences can illuminate and explain the overall bias of the entire article. This is a difficult task, however, considering that ideological bias tends to be implicit and subtle, and a bias sentence itself can appear merely neutral. While sophisticated semantic reasoning …

Content truncated to 500 words