Abstract: Aiming at the customer segmentation in the basic data of electric power marketing, an innovative method is proposed to achieve accurate classification of major customers based on jieba Chinese word segmentation. A self-defined dictionary containing the basic categories of customers is constructed, and jieba is used to complete the segmentation of the text data. The classification feature database is built based on the classification rules of high-frequency words and keywords in the word segmentation results. The classification feature database is used as the input of the neural network pre-training model to train the neural network model of customer classification, and the accurate classification results of power customers are finally output. This method solves the problem of unclear user category or too complicated classification method in the current power system database and provides the basis for the power company to develop differentiated customer service.

Key words: customer classification, Chinese word segmentation, jieba, neural network