Although social networks provide a platform for users to freely express their opinions, the group polarization formed by Internet users during the formation and development of social network public opinion also brings serious social, economic and political impacts.The phenomenon of understanding group polarization becomes a new problem in the development of social network public opinion.[Method/process] A data-driven user portrait model of polarized groups in social networks is constructed at three levels: data level, analysis level and conceptuallevel. A web crawler is used to collect data on the topic of“divorce cooling-off period”in Sina Weibo platform the evolution and types of polarized groups, static characteristics, dynamic characteristics, and portrait analysis of polarized groups in each period are also analyzed.[Result/conclusion] The results of data analysis show that polarized groups show a trend from multi-polarization to bipolarization during the dynamic evolution of polarized opinions. The portraits of polarized groups with different positions in each period show composite characteristics, with both the same patterns and obvious differences. This study enriches the theories and methods of group polarization and portrait research, and provides practical guidance for the intervention and guidance of online public opinion through empirical analysis.