This study employs a four-dimensional analytical framework (trigger factors

propagation pathways

threshold identification

intervention timing) to dissect public opinion crises on short-video platforms. Core findings reveal: A composite ​crisis value = platform coefficient × (emotional intensity × diffusion velocity × node density)​​ serves as the key diagnostic model

where the semantic weight of anger accounts for 0.85

and the platform coefficients for Douyin

Kuaishou

and TikTok are 1.0

0.9

and 1.2 respectively. Intervention effectiveness exhibits a three-phase decay pattern: during the ​Golden Window (0-30 minutes)​

the success rate of intervention reaches 92.7%; in the ​Critical Window (30-120 minutes)​

a 10-minute delay increases intervention costs by 36.8%; during the ​Decay Period (>120 minutes)​

the rate of secondary crisis generation reaches 83%. A monitoring system based on a ​Federated Learning architecture​ enables real-time data processing at a rate of 8.6 seconds per 10

000 entries.