To solve the problem of lack on automatic classification and annotation of a large number of violation cases in power supply enterprises, a Neural Network Language Model (NNLM) based on word vector is proposed in this paper. This model adopts word vector to represent text features, extracts essential features of text information by using neural network. Simulation studies are carried out on data collected from South China Grid to evaluate the performance of NNLM, the result of which is compared with Naive Bayes Classifier (NBC) and Logistic Regression (LR). The experimental results show that the classification accuracy of the NNLM is as high as 99%, which is much higher than classification accuracy of the NBC and LR. The NNLM effectively improves the accuracy rate of classification and the efficiency of annotation.