Research On Customer Demand Forecasting Model Based On Terminal Application And Its System Realization

With the rapid development of the Internet and communication technologies,as well as the general increase of the national consumption level,the penetration rate of smart terminals is increasing year by year.Such devices,with high performance,high integration and low power consumption,provide users with a large number of powerful application programs,so as to meet the needs of users in various fields such as work,entertainment and life.Among them,the Smartphone could not only provide basic communication services for people,but also can provide abundant choices of brand,price,appearance,and performance,which can be called the most well-known masterpiece in the intelligent terminal equipments.The coverage of mobile applications is getting better,people's dependence on Smartphone is also getting stronger,which becomes far beyond the basic communication needs now.In recent years,as a result of the rapid renovation of mobile terminal model,the continuously enhancing of the processing performance,and the acceptable price,people no longer replace the Smartphone only for manage,but often for the pursuit of better user experience.And this experience depends largely on the using perception of mobile applications running on the Smartphone.Based on this,how to analyze and predict the user's replacement needs are mainly discussed in this thesis according to the use of mobile application.The application usage of the user's mobile needs to be tapped from the Internet log.Meanwhile,the communication operator not only has the natural advantages of obtaining these logs,but also has the motivation to analyze the potential replacement requirement of users.This will open a new prospect in the mobile terminal sales market,so as to keep the existing users and developing new users in the competition with other operators.With the data sources provided by communication operators,the status quo of mobile data mining research is analyzed first in order to seek the reference and basis of data mining.After that,how to extract and analyze user's Internet logs for data mining are introduced,as well as the log size,the main fields used for analysis and the statistical characteristics of log data.Then,the Temporal Bag-of-Apps data model is proposed in this thesis,which splits the 24 hours of a day into four time segments and then maps the application usage into four time segments to build user log attributes and data modeling.The rationality of the modeling is analyzed and verified,and the user behavior characteristics are briefly analyzed.Based on these analyses,two methods to construct the predictive model of the Replacement behavior are studies.The first one is the unsupervised filtering algorithm based on the splitter's result.The second one is the survival analysis method based on the Cox's risk model.Finally,the performances of the two prediction models are presented and analyzed separately.According to the results,the risk model has better comprehensive prediction accuracy.