Research On Prediction Algorithm Of Web Customer Churn

The rapid development of the Internet and the advent of the era of big data, make enterprise to change marketing model, not only the traditional offline mode but also develop network marketing, which is called e-commerce. Online shopping access to a larger number of customers by its convenience, but the convenience make the platform user more unstable as the same time, namely customer churn rate is very high. The improvement of corporate earnings can be get by developing new customers, but the development of new customers often need to pay a greater price. Relatively speaking, retain old customers cost less and get more gains. Therefore, e-commerce enterprises need to effectively manage users to reduce the customer churn rate and to occupy the dominant position in the fierce competition and have a relatively good development. There are significant difference between web customers and telecommunications and financial industries. Telecommunications and financial industry customers belong to contract customers, these customers churn can easily defined. But, the web customers, such as e-commerce customers, belongs to the typical non-contractual customer, it is difficult to predict customer state and customer churn rate is relatively higher. The research of customer chum problem in telecommunications and financial industries both at home and abroad early started and have been got abundant research results. At present, the research of web customer chum prediction is less at home and abroad, mainly used research methods in these researches including traditional statistical methods, statistical machine learning methods and self-organization method. This article research web customer churn prediction from two aspects, customer value and customer reviews emotion. The main research work is as follows:1. In this paper, on the basis of the customer life cycle theory, in combination with the practical business needs of the network shopping, we put forward a kind definition of network loss customer. And distinguish the loss of customers from customers according to the definition. Then, based on customer value theory, we select the three attributes in RFM model and other Derived Attributes based on RFM, as the main properties of construct prediction model. Finally, we get five attributes:the customer buy time for the first time, the customer buy time recently, customer purchase frequency, customer purchase amount and customer score. Through verification and comparison on various predictive classification algorithm, we found that feature selection based on customer value has good prediction effect, and the definition of network loss customer that we give based on the customer life cycle definition can accurately distinguish the category of customers.2. Web customers can not face to face communication with enterprise in the shopping process, and the channels that enterprise obtain customer feedback are relatively scarce. Customer comments as an important way of enterprise to receive customer feedback plays an important role on improve enterprises service and maintain customer. Considering the influence the customer comment text show in the process of customer shopping and the impact the customer personal emotional tendency generated on loss of customers, in this paper, we integrate the customer comment emotion into model built on customer value theory, as a new attribute of forecasting model, combined with the first five attributes, there are six main attributes in the new model. To test and verify the prediction effect of the new model, we make contrast test in a variety of commonly used prediction and classification algorithms, such as SVM, neural network, test the new model prediction effect. The experimental results show that the performance of new model is better than the old model. Therefore, the customer comments emotion is an important factor of customer churn. It can effectively promote model prediction effect.