Discovering cross-platform narrative flow templates using frequent subgraph mining

Social media narratives often transcend platform boundaries, yet existing models rarely capture the evolution of ideas across them. This paper presents a graph-mining framework for discovering narrative flow templates–recurrent structural patterns that describe how narratives propagate between platforms over time. We adapt the SoPaGraMi algorithm, originally developed for frequent subgraph mining, to operate on a unified narrative flow graph where nodes represent platform-specific narrative instances and directed edges encode temporal relationships. Using the discourse surrounding the recent Tariff War as a case study, we analyze a cross-platform narrative graph drawn from 19,303 YouTube (Y), 11,218 TikTok (T), 12,252 X (X), and 30,493 Instagram (I) posts. The approach uncovers distinct diffusion motifs, including the full-chain template $$Y \!\rightarrow \! T \!\rightarrow \! X \!\rightarrow \! I$$ – a pro-American narrative flow template, $$Y \!\rightarrow \! I \!\rightarrow \! Y$$ - a pro-Chinese narrative flow template, and $$I \!\rightarrow \! Y \!\rightarrow \! I$$ reflecting cooperative U.S.–China narratives. These findings reveal how different platforms serve complementary analytical, expressive, and visual functions. Our results demonstrate that frequent subgraph mining can identify coherent cross-platform diffusion structures, offering a scalable foundation for understanding narrative diffusion in multimodal social ecosystems.