A Computational Analysis of Social Media Reframing of News Events

Achyutarama R Ganti, Steven R Wilson

Keywords: framing analysis, reframing on social media, cross-domain analysis, large language models, computational framing

Abstract:

Framing refers to the process by which certain aspects of reality are selectively emphasized
in order to promote particular interpretations of public issues. Frames influence not only
what audiences attend to, but also how and who they assign responsibility to, evaluate policy
solutions, and form political attitudes. Reframing happens when existing frames are
contested, reinterpreted or altered as information moves through platforms. While traditional
news usually introduces particular frames, social media provides space for people to reinterpret,
amplify or challenge these frames through their discussions. Nearly half of Americans turn to
social media as a regular source of news, and what they read helps them to form opinions
about what is happening in the world. In fact, U.S. adults under 30 trust information from
social media almost as much as information from national news outlets As we notice a vast
amount of digital text across news and social media platforms these days, understanding how
issues are framed becomes important. The scale at which information is circulated makes it
necessary to develop approaches that can identify and analyze framing patters across large
volumes of data.
Computational approaches have been used to scale framing detection across large corpora.
However, much of this work remains limited to single domains or single platforms
which makes it difficult to compare how framing patterns differ across issues and platforms.
While prior work has explored document-level framing, Little work has been done to ex-
amine reframing of events.
So, we perform a cross-domain investigation of framing and reframing across three soci-
etally important issues: immigration, the environment, and public health. We adopt a unified
taxonomy grounded in prior literature and apply it across the domains. This allows to
compare framing patterns across issues and provides us a framework to evaluate both human
annotations and model predictions. In our study, we address three research questions.
RQ1: How do framing patterns vary across major public issue domains when analyzed
using a shared taxonomy?
RQ2: How do frames introduced in news coverage become reframed as they circulate
through social media discussions?
RQ3: How closely do large language model predictions align with human framing annota-
tions?

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 96