The fragmentation of youth wellness discourse across diverse social media platforms poses significant challenges for public health monitoring. While distinct communities exist on Reddit, TikTok, and YouTube, the structural and semantic bridges that facilitate the flow of harmful content remain understudied. In this work, we present a cross-platform analysis of youth wellness ecosystems using a Social-Semantic Knowledge Graph (SSKG) approach. By integrating data from 10,396 users across three platforms on multimodal data, we map the interplay between social interactions and semantic entities. Our analysis reveals that 14.9% of users engage in "high-risk" echo chambers characterized by high cohesion (>0.65) and semantic isolation. Crucially, we identify "bridge nodes" that serve as transit points connecting healthy fitness communities to harmful subgroups. To support scalable and interpretable intervention, we implement a LightRAG-augmented retrieval system that mitigates the hallucinations of standard LLMs, allowing researchers to trace these risks back to specific evidence. This work contributes both a rigorous methodological pipeline and critical empirical insights into the cross-platform nature of online vulnerability.

PDF