To meet the needs of intelligent government application, an intelligent e-government system based on text mining is developed, and its performance is evaluated experimentally. The tasks of the system include the realization of the correct classification of the public's question messages, the acquisition of the message hot spots and hot words, and the automatic scoring and evaluation on the quality of the message replies. The effectiveness and efficiency of support vector machines, logistic regression, random forest and naive Bayes in classifying public comments were compared and analyzed. A method for hot spot problem mining based on clustering and linear discriminant analysis is proposed. The rules for evaluating the quality of message responses based on relevance, integrity and interpretability are defined. Based on the data set provided by the eighth "Teddy cup" data mining challenge in China in 2020, the experiment and test results show that the classification accuracy and recall rate of messages are both more than 84%, which can effectively screen hot issues.