Abstract: Based on the value cognition of consumers, it is a critical approach for telecom operators to classify consumers correctly in order to know them better. Telecom operators can divide the consumers into different categories and develop different service policies as well, so that they can advance the differentiated marketing to improve business benefits. Firstly, consumers of different telecom networks are classified in detail. In order to improve the classification accuracy, this research improves the determination of learning rate and the initial value's weight vectors based on the traditional self-organizing map neural network, proposing an improved self-organizing map neural network (ISOMNN). Then this ISOMNN is applied to the simulation experiment by using the consumer data provided by a telecom operator of a province. The simulation results reveal that ISOMNN can not only take the stability into account, but also solve the slow self-organizing problem well, boosting the accuracy of consumer-classification and reducing errors significantly. Finally, based on the results of classification, this paper provides the basic principles for the telecom operators, which will help them to implement differential marketing.

Key words: improved self-organizing map, different telecom network consumers, subdivision research