New Research Unifies Models of Online Intergroup Hostility

What changed

Researchers have developed a unified empirical framework to model the mechanisms of six foundational theories of intergroup hostility: boundary construction, threat construction, scapegoating, negative evaluation, dehumanization, and action orientation. This work analyzes 2.86 million posts from TikTok, Truth Social, and Twitter/X collected during the 2024 U.S. presidential election. The study aims to bridge the gap between parallel-developed theories in social psychology, moral psychology, and political science, which often propose conflicting accounts of hostility development and have rarely been tested against each other in real discourse. By integrating these theories into a common framework, the research seeks to provide a more comprehensive understanding of how intergroup hostility rhetoric is organized and manifests online.

The analysis reveals a robust empirical system where boundary construction, threat construction, and action orientation form the core. Structurally, boundary construction and threat construction anchor the system. Temporally, a regular ordering of mechanisms was observed: boundary construction, negative evaluation, and action orientation tend to appear early in discourse. Dehumanization and threat construction emerge later, while scapegoating appears latest. This temporal sequencing suggests a developmental pattern in how hostile rhetoric unfolds online.

Why it matters for builders

This research offers AI builders a more nuanced and empirically grounded understanding of intergroup hostility rhetoric. By unifying disparate theoretical models into a single framework, it provides a clearer basis for developing computational social science tools. This can lead to more effective moderation systems that can identify and address various forms of harmful speech, moving beyond single-label detection. The identified temporal ordering of mechanisms could also enable earlier detection of escalating hostility.

Practical impact

AI builders can leverage this research to enhance natural language processing models for content moderation. The unified framework and identified mechanisms can inform the design of classifiers that are more sensitive to the subtle progression of hostile rhetoric. For instance, models could be trained to recognize early indicators like boundary construction and negative evaluation, allowing for intervention before more severe forms like dehumanization or scapegoating emerge. The study's findings can also guide the development of tools for analyzing discourse patterns across different platforms, aiding in cross-platform moderation strategies and research into polarization.

Caveats and source limits

The study relies on data from the 2024 U.S. presidential election and three specific social media platforms (TikTok, Truth Social, Twitter/X). While the findings offer a unified model, the generalizability to other contexts, time periods, or platforms may require further investigation. The research focuses on modeling the mechanisms of hostility rhetoric; the effectiveness and ethical implications of moderation strategies based on these models are not detailed within the provided excerpt. The code and data are available via a tinyurl link, but direct access and examination of implementation details are not provided here.