A Study On The Statistical Methods Of User Images In The Background Of Big Data

Since 21 th century, with the continuous development and renovation of technology of internet and mobile, mobile internet, which is the product of the two, rapidly develops in these years. According to data display, only in 2014 the smartphone quantity in China reached 780 million units, and had 570 million mobile internet users, these number is increasing with very quick speed. As the length and frequency people use smart phone increased, the user’s data and behavior data are exponentially increased. Besides, collecting data from intelligent terminal, which has basic characteristics of large amount, real-time, accuracy, space and dynamic. In order to solve this big data need, This article collected behavior data of using cellular phone applications in the fourth quarter of 2014 from a set of users of smartphone, let users be the study subject, analyze and practice the three problems of user profile, user loss model prediction, clustering of user behavior.First of all, research suggests that user profile is mathematical modeling for users in the real world. The core of user profile is the establishment of tag system. Label is a kind symbol of user characteristics, user profile could express by a set of tags. Secondly, based on user behavior data over a period, user loss prediction model is established trough two methods of analysis-the support vector machine and survival analysis. Model results show that, the accuracy for user erosion prediction can reach more than 90 %, recall rate is higher than 80%. Finally, using the mahout analysis framework in hadoop, using more than 20 user behavior indicators to do clustering analysis, then got preference and habits characteristic differ from classes of user groups when they using apps. And extract a certain user behavior data, the cluster analysis considers that users can be divided in six groups, and gives corresponding retention strategy and marketing advice according different groups.This paper sums up and summarizes the definition of user profile, analyzes basis process and statistical analysis methods to establish user label system. Proposed three basic elements for the study of user profile, which, represents the user’s attributes, records users’ behavior through their life, describes user loss. Refer to definition of loss, make new definition based on the actual scene for app users’ loss, establish model to predict user loss behavior. For user profile in the user behavior research, this paper will apply FRM indicators in marketing to analyze users’ behavior, combined with statistical method of cluster analysis, better explains and describes characteristics of user behavior.