Application Research Of Data Mining Technology In Customer Relationship Management

The rational use of customer relationship management is one of the key factors for business success.The analysis of customer relationship management is based on customer relationship management,through the use of information technology means to operate the company's daily business data,and then help companies make better decisions.Analytical customer relationship management has many aspects.In this paper,from the most important in the two aspects of customer segmentation and customer identification.In terms of customer segmentation,many studies have conducted clustering operations in the traditional RFM analysis model.However,there are few features in the RFM model and customers cannot be measured from the whole.Therefore,aiming at this problem,starting from the actual business requirements,a multi-index customer analysis model is proposed based on the RFM analysis model,and use the entropy method to adjust the feature weights which makes the customer's feature extraction more effective.According to the clustering results,it is found that the multi index analysis model is more compact than the RFM model,and it is more reasonable for the division of customer grade.In customer identification,the core is to establish a prediction model.At present,most of the traditional data mining algorithms are used to build prediction models.Although some results have been achieved,most of these algorithms belong to shallow learning.There are few nonlinear transfer layers,which may lead to under fitting phenomenon and the prediction results are not accurate.Therefore,deep learning is introduced in this paper,which can improve the under fitting phenomenon and improve the prediction effect by building complex hierarchical structure.This paper focuses on the DNN and CNN algorithm in deep learning,and establishes the prediction model.The DNN prediction model sets the number of different hidden layers by experiment,so as to determine the model structure.For the gradient disappearance phenomenon,first use two kinds of activation functions to compare experiments,and then find a better performing function to improve the phenomenon,and adopt dropout method to avoid over fitting phenomenon in full connection to realize model optimization.The CNN prediction model designs four structures by adjusting the convolution layer,the pool layer and the convolution kernel.The weights initialization method is used in the parameter adjustment,and it is combined with the stochastic gradient descent method and the momentum method to realize the model optimization.Finally,the two best prediction model is applied to realize the function of a business customer identification system,found that build prediction model based on the deep learning performance are higher than the traditional data mining algorithms to construct model.Among them,the prediction model optimized by CNN is the best.