跳到正文
arXiv cs.CY· Upasana Dutta, Homa Hosseinmardi, Amir Ghasemian, Aaron Clauset, Duncan J. Watts·· 9 小时前AI 评分34

研究:不对称修辞策略驱动美国电视新闻的党派倾向

Asymmetric rhetorical strategies drive partisanship in television news

AI 导读

一项用 LLM 分析 2013–2025 年超 40 万期新闻节目的研究,识别出 8 个可解释语言特征,解释了美国六大电视台党派倾向 73% 的方差。MSNBC 和 CNN 在共和党总统任期内左移、民主党任期内趋于温和,而 Fox News 每次大选后保守倾向都增强,且其党内赞扬与外党攻击的关联强度是 MSNBC 的三倍。

正文 · 原文

View PDF HTML (experimental)

Abstract:Television remains a dominant source of political information in the United States, yet systematic evidence on how partisanship is expressed in televised news remains limited. Prior research has largely treated partisanship as a one-dimensional construct, offering limited insight into the rhetorical strategies that underlie ideological bias. Here, by analyzing over 400,000 news episodes using large language models (LLMs), we provide a multi-dimensional measure of partisanship bias in television news across six major U.S. news stations (2013--2025). We identify eight interpretable linguistic features that explain 73% of partisanship variance in television news.\ Our findings reveal systematic asymmetries across stations. While MSNBC and CNN shifted leftward during Republican presidencies and moderated under Democratic presidencies, Fox News increased its conservative alignment following each election regardless of the electoral outcome. Beyond aggregate trends, MSNBC and CNN emphasized attacks on Republican actors more than reinforcement of Democratic positions, whereas Fox News reinforced Republican viewpoints substantially more than it attacked Democratic viewpoints. The association between in-group praise and out-group attacks was also threefold stronger on Fox News than on MSNBC. These results are validated against trained human raters and replicated with four additional LLMs with differing political leanings. Together, our study reveals systematic station-level differences in the rhetorical strategies with which ideological slant is communicated in television news, opening new opportunities for future research on media effects, audience attitudes, and political communication.
Comments: 39 pages (15 main text + 24 supplementary information), 7 figures, 1 table; 24 supplementary figures, 4 supplementary tables
Subjects: Computers and Society (cs.CY)
Cite as: arXiv:2611.25140 [cs.CY]
  (or arXiv:2611.25140v1 [cs.CY] for this version)
  https://doi.org/10.48550/arXiv.2611.25140

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Upasana Dutta [view email]
[v1] Wed, 18 Nov 2026 20:19:19 UTC (10,377 KB)

来源:arXiv cs.CY · arxiv.org