Multimodal Information Risk Flow in Short-Video Crisis Environments

Public emergencies are increasingly shared and discussed on short-video platforms, where videos, user interactions, and timing influence how social risks become visible and spread. This study examines a major short-video platform in China and proposes a Multimodal Information Risk Flow (MIRF) framework to understand how risk signals appear, are categorized, and evolve over time during emergencies. We analysed 250,100 anonymized video–comment records collected between 2022 and 2024, combining text, images, audio, user behavior, and author information to study patterns of risk representation and amplification. Our results show that social risk spreads unevenly over time and is strongly influenced by the type of content, with toxic comments leading to faster and larger cascades. General engagement metrics play a smaller role, and shorter times to peak activity are consistently linked to larger risk cascades. These findings highlight that social risk on short-video platforms is multimodal and temporally concentrated, and caution is needed when generalizing results beyond similar platforms and regulatory contexts.