基于价值嵌入与JS散度的舆情主体—行为价值观冲突检测模型研究OACHSSCD

[Purpose]The underlying cause of public opinion outbreaks may stem from implicit conflicts between the identity of the sub-jects involved in an event and the values conveyed by their behaviors.[Method]This paper proposes a subject-behavior value conflict detection model based on value embedding and semantic conflict quantification.First,by integrating sociological value theory with large language models,we construct an unsupervised,multi-domain value lexicon.This lexicon is embedded into a high-dimensional semantic space to form value prototype vectors.Subsequently,domain-adapted prompt templates are designed to quantify the subjects and behaviors within public opinion events as probability distributions over this value prototype space.Finally,leveraging the Jensen-Shannon diver-gence from information theory,a value conflict intensity function is constructed to measure the significance of differences between these two probability distributions.[Result/Conclusion]This study provides a novel theoretical perspective and a computational paradigm for public opinion analysis,holding substantial theoretical significance and practical value for the precise early warning and in-depth investiga-tion of major public opinion events.