Bridging the Narrative Divide: Cross-Platform Discourse Networks in Fragmented Ecosystems

TL;DR

This work tackles the fragmentation of political discourse across social platforms by proposing a platform-agnostic discourse network, CANE with its temporal variant t-CANE, that connects users through latent narratives rather than platform-specific interactions. By embedding content, clustering into semantic narratives with DP-Means, and constructing cluster-affiliation–based graphs, the approach yields scalable, cross-platform user graphs that support information-operation detection, ideological mapping, and cross-platform engagement prediction with high data efficiency. The analysis of X and Truth Social reveals a small set of bridge users who structurally enable the majority of cross-platform narrative diffusion, forming a bridge zone that standard interaction- or lexical-based graphs fail to uncover. The framework offers a scalable lens on narrative diffusion with implications for governance, moderation, and policy interventions in fragmented information ecosystems, while acknowledging limitations and outlining ethical considerations for future work across additional platforms.

Abstract

Political discourse has grown increasingly fragmented across different social platforms, making it challenging to trace how narratives spread and evolve within such a fragmented information ecosystem. Reconstructing social graphs and information diffusion networks is challenging, and available strategies typically depend on platform-specific features and behavioral signals which are often incompatible across systems and increasingly restricted. To address these challenges, we present a platform-agnostic framework that allows to accurately and efficiently reconstruct the underlying social graph of users' cross-platform interactions, based on discovering latent narratives and users' participation therein. Our method achieves state-of-the-art performance in key network-based tasks: information operation detection, ideological stance prediction, and cross-platform engagement prediction$\unicode{x2013}$$\unicode{x2013}$while requiring significantly less data than existing alternatives and capturing a broader set of users. When applied to cross-platform information dynamics between Truth Social and X (formerly Twitter), our framework reveals a small, mixed-platform group of $\textit{bridge users}$, comprising just 0.33% of users and 2.14% of posts, who introduce nearly 70% of $\textit{migrating narratives}$ to the receiving platform. These findings offer a structural lens for anticipating how narratives traverse fragmented information ecosystems, with implications for cross-platform governance, content moderation, and policy interventions.

Figures (5)

• Figure 1: Overview of our cross-platform user network inference framework. Content is embedded and clustered into semantically coherent narratives, which form the basis for constructing user-user networks. These networks support a range of downstream tasks, including community detection, stance prediction, and content analysis.
• Figure 2: A conceptual illustration of cross-platform narrative diffusion. While most communities are siloed within platforms, some users form bridge zones: structural overlaps where narrative transfer is more likely.
• Figure 3: Comparison of computational complexity across network construction methods using empirically-informed scaling trends. CANE and t-CANE scale more efficiently with users due to FAISS and clustering, while baselines exhibit steep cost increases tied to post or user volume.
• Figure 4: Proportion of peak AUC achieved as a function of training data percentage. Each curve is truncated at the first point where the method reaches 95% or more of its peak performance, illustrating the data required to achieve near-optimal results.
• Figure 5: Visualization of the discourse network colored by platform (blue: X, red: Truth Social), with node size scaled by degree. This layout reveals a structurally embedded bridge zone: a dense, mixed-platform region near the center where users from both platforms are highly interconnected. These users serve as key conduits for narrative migration across fragmented media environments. The concentration of red nodes within the blue-majority core illustrates cross-platform entanglement not evident in interaction-based graphs.