Traditional methods extract users'static behavior characteristics and use supervised learning model to complete the identification. Under the condition of large-scale social network, the improper behavior characteristics of Navy user team are more and more similar to normal users, so it is impossible to accurately identify the improper behavior of Navy user team in social network. In order to solve this problem, this paper studies the accurate identification technology of improper behavior of naval user team in social networks based on the dynamic behavior characteristics of users. Based on the description of social network, the dynamic characteristics of user behavior are extracted. In this paper, the problem of identifying the improper behavior of Navy user team in social network is regarded as a two-class problem. The corresponding samples of dynamic features are taken as input, and a decision tree is constructed to identify the improper behavior of Navy user team in the new social network data set. The results show that the dynamic features can effectively reflect the misconduct behavior of the naval user team; the recognition results of the proposed technology are basically consistent with those of the manual identification; the harmonic mean and balance accuracy of the proposed technology on the three data sets are higher than those of other technologies. It can be seen that the accuracy of the proposed technology is high.