Multi-Modal Framing Analysis of News
Summary
This paper argues that computational framing analysis of news has been artificially constrained by two assumptions: that frames live only in text, and that they can be drawn from small predefined inventories. The authors propose a multi-modal framework that jointly analyzes textual, visual, and text-image cues using open discourse-analytic framing categories rather than fixed frame sets. Applied to a corpus of U.S. gun violence coverage, the approach uncovers systematic differences in how outlets across the political spectrum frame the same events through their combined editorial choices of words and images.
Key Contributions
- A multi-modal framing analysis framework that integrates text, images, and their interaction, rather than treating text alone.
- A methodological shift away from fixed frame typologies toward open, discourse-analytic framing categories operationalized computationally.
- A paired text-image corpus of U.S. gun violence news coverage as a case study and resource.
- An empirical demonstration that multi-modal framing reveals political-leaning differences invisible to text-only analysis.
Methods
- Built a corpus of U.S. news articles on gun violence paired with their accompanying images.
- Automated frame extraction performed separately on text, on images, and on text-image combinations.
- Frames are drawn from discourse-analytic framing categories rather than a closed predefined set.
- Comparative analysis across outlets of differing political orientation to surface framing differences.
Findings
- Frames can be reliably extracted at textual, visual, and cross-modal levels with the proposed approach.
- Outlets show distinct multi-modal framing patterns on gun violence that correlate with political leaning.
- Visual and cross-modal cues carry editorial framing signals that purely textual methods miss.
- Editorial choices manifest not just in word selection but in image selection and text-image pairing.
Connections
No other papers have been provided under shared topics, so there are no in-corpus wikilinks to make here. The work connects intellectually to the broader literatures on framing theory in political communication, computational frame classification in NLP, and emerging multi-modal approaches to media analysis, but none of those are represented among the supplied bibtex keys.