Information overload and cocoon effect make the phenomenon of emotional polarization easily appear in online knowledge community. The mechanism of emotional polarization of users in knowledge community is analyzed, so as to reveal the formation rule of users’ emotional polarization in knowledge community and summarize the intervention measures, provide a theoretical basis for further effective control of polarized emotions of knowledge community users. Starting from the theory of social network structure, based on the PAD emotional model, starting from the dimensions of Pleasure, Arousal and Dominance, focusing on the degree of user centrality, an index framework of emotional polarization is constructed around three behavioral patterns of information retrieval, information selection and information interaction. SOR model is used to dynamically explore the polarization mechanism under this framework. The results show that the heterogeneity of social network structure has an effect on participants’ emotional perception and information behavior. The polarization of Pleasure has a positive effect on user centrality, and the polarization of Arousal and Dominance have a negative effect. User centrality positively affects their information selection and interaction behavior, but has no significant effect on retrieval behavior. The emotional polarization in the process of community knowledge sharing is decomposed from different perspectives, and the polarization mechanism is shared by combining social network structure and information behavior. From the perspective of application, this is conducive to promoting knowledge sharing, communication learning and information value chain remodeling, and also provides a kind of insightful analysis paradigm for this field.