Customer Segments In The Telecom Ring Tones Marketing

This paper starts from the PAS CRBT customers of a telecom enterprise, I introduced the research background and the applications of customer segmentation in telecom firstly, secondly I studied many researches of other people in the field of customer segmentation, summarized and compared the different methods of segmentation, then I determined which method and mode I will use in this paper according to the purpose of the research. In the real case study of this paper, after analyzed what kind of data be needed in this research, I prepared all data in order to use the different forms of the data in different models; then I set up a model using the method of cluster analysis based on the attributes of the customers' behavior, and it results several groups that have much different in the customers' behavior from each other, then I described the character of these groups and selected one of these groups to do the next analysis process. Next, I chose smaller attributes of behavior and inserted in more customers' background attributes, clustered the specific data for the second time in order to analyze what kind of attributes impact the customers'cost on CRBT, then decided what variables will be chosen to set up the prediction model. In the final step, I used the attributes what impact the cost on CRBT much and the customers'background attributes to set up the prediction model, we got the rules that what kind of people would like to use the CRBT and what kind of people would not like to use the CRBT in telecom. Finally, the model's validity is evaluated both by the test data and the real data.This paper analyzes the real data of a telecom enterprise using the data mining theory and statistical analysis method. With the rapid development and more applications of the data mining, there are more and more scholars studied the customer segmentation in telecom or bank fields using the data mining theory and algorithm, however, the result of literature retrieval shows that there are few studies focus the CRBT customers as their target, or the CRBT customers just be mentioned while studying the value-added services, an there are much fewer studies using the results of data mining to direct the CRBT marketing. the CRBT service, as a second fastest developing business in value-added services after the message service, can raise the value of consumers who become a new CRBT user by 5 Yuan, many new CRBT users can bring in much profit, so the author of this paper select the CRBT customers to do the research.On the issue of what kind of method will be selected, the author carried out the research on the successful experience of others at the same time considering the purpose of this paper. When I clustered the base data for the first time, I chose the self-organizing neural network algorithm and reduced the dimensions by factor analysis first, then set up the clustering model based on the consumers' behavior, a problem will be solved on this stage is that "what kind of people we will study?"; then I set up the clustering model by k-means fast cluster algorithm for the second time,and solved the problem of determining the value of k in the k-means algorithm through statistical analysis method, as well as the cluster initial centers, the main issue on this stage is that "what kind of people will spend more money on CRBT?" and "what kind of people will spend less money on CRBT?",with the purpose of analyzing the attributes which can impact the customers'cost on CRBT, then I used these attributes to join in the next process; finally, two different kind of decision tree model are set up based on a mixed training data composed of two parts:the CRBT consumers and the non-CRBT consumers, and the better one is chosen by comparing their prediction results,the decision tree model can result a name list which can be used to direct the marketing process, the main problem on this stage is that "what kind of people will become a new CRBT user?" and "what kind of people will not be a CRBT user?". In the last part of this paper, the prediction model was estimated from two aspects, on the one hand, I used the test data set which is similar to the training data set to test the prediction model and calculate the accuracy rate and coverage rate, adjust the parameters of the model if the rate is not satisfied; on the other hand, I used all of the non-CRBT consumers data set to test the prediction model in a real environment, then track these customers whose name are in the prediction list for six months, calculate the rate that how many people become new CRBT users naturally by month in the next six month (attention:not the result of active marketing). Then we compared to the rate of all the non-CRBT consumers and estimated the validity of the prediction model.The idea of this paper can simply be summarized like this:finding out the potential regulation through studying "what kind of people, and get the conclusion:"what kind of "people have "what kind of" behavior then they can be a potential new CRBT user.The first "what kind of refers to the positioning of the CRBT consumers,the second "what kind of "refers to the CRBT consumers'background just like the customers' age and agent,the third "what kind of refers to the behavior of the customers.The mode and methods of this paper can also be used to direct the marketing action in the other value-added services in telecom enterprises, such as the CRBT,message marketing in the CDMA users, MMS,GPRS monthly in the mobile M-Zone users,these customers who are using these business are related to their behavior and background information more or less,and the data mining technology can find out these relations and the potential regulation,so the marketing in these value-added services is similar to the CRBT marketing in this paper.It is supposed that the research of this paper explored a new mode and methods in the marketing of value-added services, and the evaluation results proved that the customer segmentation technique, working out the problem of what kind of features impact the customers'cost on CRBT much and using these attributes to set up the prediction model, is effective and practicable.