The Application Of Data Mining In Channel Preference User Identification

Under the new situation of mobile Internet,more and more telecom operators began to realize the multi-directional development of physical channels and electronic channels to provide consumers with more convenient services.At the same time,in the face of fierce competition,the channels need to rely on the precise positioning for the preferences of users to meet customer needs,realize customer diversion and implementation of differentiated marketing and services.Finally,by combining with product and channel adaptation to achieve product precision marketing,and solve the operator "wide net,low return"problem.In order to solve this problem,this paper is based on the idea ofaccurate marketing:first,find out the channel’s preferred users,and then at the right time to recommend the appropriate product,aimed at improving the effectiveness of the channel activities and customer penetration.Among them,the article focuses on how to find the various channels of the user preferences of this problem.In the implementation process,in order to reduce the complexity of the model,we will be based on the current mainstream channels were identified by the user preference model.Based on the enterprise data warehouse and subordinate data mart,this paper makes use of the light summary data of all kinds of production systems to make up for the drawbacks of narrow data surface and short data period.The user’s preference information and behaviors are extracted,and the user’s preference degree of the channel is forecasted by using logistic regression and random forest data mining algorithm respectively.The whole mining process includes the following three aspects:business objective,variable design,data preparation,variable selection,data preprocessing and model building and evaluation.After the business problem is solved according to business requirement,data mining is carried out.Design,wide-table construction,and then SQL access results for data quality inspection and variable exploration,to determine the need to retain the relative "important" variables,and finally through constant adjustment of parameters and model fitting,different algorithms produce model effects To select the optimal model for curing.Based on TWM and R mining tools,this paper constructs the model using Logistic regression and stochastic forest algorithm respectively,and uses the model results and business rules to validate the target users.The results show that both can achieve higher coverage and promotion degree.After synthetically considering the model effect and the curing cost,this paper chooses the logistic regression algorithm and realizes the effect in the library.On the other hand,according to the forecasting probability,the users of the whole network are divided into several sub-files,and the marketing test of the users is carried out by stratified sampling.The test results show that the model is very effective.In order to better realize the combination of channels and products,this paper proposes to achieve the user-channel through the channel of the user identification model derived from other channels,according to the product in different channels of the marketing results indicators to achieve product and channel adaptation,-Product of the precision marketing model.