Abstract: 【Purpose /significance】In order to grasp the basic characteristics and development law of network public opinion in the en⁃rollment examination and find out the possible public opinion in time, the study makes an evolutionary study on the relevant topic texts of domestic mainstream network social platforms during the re examination of postgraduate entrance examination.【Method /process】Collect data with Python, model word pairs with BTM model, mine and classify topics. Analyze the strength and content evolution of each topic over time【. Result /conclusion】The results of theme intensity and content evolution show that public concerns are closely re⁃lated to the process of enrollment, and show a certain cycle and regularity, which can be used as the basis for network public opinion prediction.【Innovation/limitation】BTM model solves the problem of data sparsity in the corpus of short articles and can effectively mine topics. However, it also has the problem of insufficient semantic understanding of machine learning and manual interpretation, which needs to be improved in follow-up research.