Unsupervised framing analysis for social media discourse in polarizing events

Summary

This paper proposes an unsupervised methodological pipeline for surfacing frames in polarized social media discourse without relying on predefined frame inventories. Combining machine learning, network analysis, and natural language processing, the authors argue that emergent and subtle framing structures in online conversations can be made visible at scale. The contribution is primarily methodological: bridging classical framing theory in communication research with computational social science techniques to study how polarization manifests discursively on platforms.

Key Contributions

• An integrated, unsupervised pipeline for frame detection that does not require pre-labeled frame categories.
• A methodological bridge between framing theory and computational tools (ML + NLP + network analysis).
• A conceptual extension of how emergent frames can be identified and characterized in polarized online discussions.
• A scalable approach suited to the messy, evolving vocabulary of social media discourse around contentious events.

Methods

• Unsupervised analytical methodology designed to avoid imposing a priori frame taxonomies.
• Machine learning components to cluster or represent textual content.
• NLP tools for processing and characterizing the language of frames.
• Network analysis algorithms to model relations among frames, terms, or actors within polarized conversations.

Findings

• The abstract reports no specific empirical results; the paper’s contribution is the methodological pipeline itself and its proposed analytical affordances.
• The framework is presented as capable of revealing latent, emergent framing patterns that supervised or dictionary-based approaches would miss.

Connections

This work sits alongside other computational-social-science methods papers that develop unsupervised or semi-supervised pipelines for analyzing contentious online discourse, such as Scalco2026-bd and Balluff2026-if. Its focus on frames in polarizing events connects to broader empirical studies of polarization dynamics like Gaisbauer2025-by and Esau2025-tf, and to work on how partisan or hyperpartisan discourse structures itself on platforms, e.g., Bouchaud2026-lr and Rossini2026-jn.