Panoramic Persona And Model Prediction For Enterprise Marketing

With the continuous development of network and information technology,the data continue to accumulate on the basis of business processing,we enter the Data Technology era from the Information Technology era.Enterprise marketing methods are also from the Product,Price,Place,Promotion this 4P theory to Consumer,Cost,Convenience,Communication this 4C theory,user-centered precision marketing is business needs.But now the enterprise's understanding of the user is not clear,the user information is incomplete,in order to improve the user's understanding of the user,in this thesis,we focus on the panoramic persona and model prediction,combined with KTV Online to Offline actual scene,and finally realize the refinement of the enterprise operations.The main work of this thesis includes the following aspects:Firstly,this thesis designs a distributed processing framework.In this thesis,the Hadoop distributed file system and Hive are used to realize the data storage and management.The Impala system is used to build the user's image.The model is predicted by the Spark cluster.Finally,the distributed data storage,management and analysis are realized.Secondly,this thesis realizes the construction of persona based on multi-source data fusion.In this thesis,persona is designed from internal and external data,multi-dimensional business data,multi-directional attribute granularity,persona is achieved through the Impala SQL direct access,statistical transformation,natural language processing,regular matching,rule determination,user event model,etc.and ultimately achieve user understanding and meet the corporate marketing business.Thirdly,in this thesis,a hybrid model of gradient boosting decision tree and linear model fusion is realized.The method uses gradient boosting decision tree to realize feature discovery automatically,and uses tree path to extend eigenvector and linear model to improve model precision.In this thesis,the method is applied to user gender classification and user spending prediction model,and a variety of schemes including random forest and gradient boosting decision tree are designed to verify the validity and accuracy of the scheme.