The invention provides an electric power overhaul text mining method based on a deep neural network. The electric power overhaul text mining method comprises the steps of obtaining an electric power overhaul text generated by an electric power service hotline service system; Carrying out processing of converting the text data into numerical data on the obtained electric power overhaul text; And performing semantic classification based on a neural network on the processed data to obtain a classification result indicating a specific demand. By introducing a universal large-scale pre-training word vector, the classification effect is not influenced by reduction of professional word coverage, and on the contrary, the classification accuracy is greatly improved. The introduction of large-scaleexternal word vectors has important significance for improving the effect of complex text mining tasks in the power industry.