To improve the efficiency of organizing and retrieving hazard information data and support more complex information processing tasks, effective technical methods need to be adopted for automatic data classification and type analysis. Support Vector Machine (SVM) can automatically classify free text. However, the working principle of the algorithm is to find the optimal classification boundary in the training set, and cannot discover typical type features. So, a normalized entropy model is proposed to search for typical type features, which improves the current TFIDF (Term Frequency Inverse Document Frequency) type feature recognition method. Taking 2534 law enforcement inspection records from a government emergency management bureau as an example, SVM was used for automatic classification, with an accuracy rate of up to 97%. At the same time, the normalized entropy model was used to provide typical characteristics of each type, providing decision support for formulating special rectification strategies for hazard investigation. The experimental results show that the combination of SVM and normalized entropy model can efficiently solve the comprehensive problem of text classification and type feature recognition.