Identifying and profiling high-risk social media information in higher education: a multidimensional risk assessment approach

Social media has become an important arena for public discourse on higher education, where information diffusion may develop into public opinion risk. This study develops a multidimensional risk assessment and information-profiling approach based on information-source attributes, content characteristics, and dissemination patterns. Using 14,911 Weibo posts from a university-related public opinion event, entropy-derived indicator weights were used to generate continuous composite risk scores. The scores were divided into minimal, low, moderate, and high relative risk levels using the 50th, 75th, and 90th percentiles, and posts above the 90th percentile were retained for information profiling. A blinded four-level coding procedure was used to assess inter-coder reliability and the association between composite scores and human judgments. LDA topic modeling characterized corpus-level semantic evolution, while K -Modes clustering identified four profiles within the upper-decile subset: low-reach emotional expression, high-reach neutral long-form commentary, high-reach negative amplification, and low-reach neutral residual discussion. The findings show that high-priority information differs in source reach, content form, sentiment polarity, and lifecycle stage. The proposed framework supports differentiated monitoring and governance-oriented interpretation, while the resulting risk levels are understood as event-specific priorities rather than universally applicable thresholds.