Objectives In this study, we develop a model based on big data analysis to find patterns in North Korean nuclear provocations. Methods Using automated text analysis classification through supervised machine learning techniques, we analyze the North Korean Central News Agency (KCNA) from 1997 to 2013. Results We find an interesting difference between the Kim Jong‐il era and the Kim and Jong‐un period, implying a major shift in the nuclear policy of Pyongyang. While Kim Jong‐il, who had a firm grip on power, focused on the international audience before conducting nuclear tests, Kim Jong‐un during his succession targeted the domestic audience prior to nuclear tests, probably in an attempt to consolidate his precarious power. Conclusion The machine learning technique allows us to analyze the effect of political communication even in authoritarian governments.