Mobile Phone Users Based On The Context Information Of Retention Time Prediction Research

Location Based Service, which is an important part of Mobile Internet Ser-vice, has a huge market scale and the interests of the good prospects. However, it still has big promotion space. Mobile service providers and manufacturers try to provide services and devices that take advantage of the location information associated with devices to provide a more personalized experience for users. For many such services, the user experience can be dramatically improved if a mobile device can predict how long a mobile user will stay at the current location.This thesis tries modeling the relevance between the stay time of mobile users and contextual information, and takes advantage of contextual informa-tion for predicting the stay time of mobile users. Specifically, the research is conducted through the following four steps.First, collect contextual mobile data through Cell Phone Information Col-lection Software deployed in mobile devices. In the experiment, we collect twenty students’cell phone data. The contextual mobile data contains time information, position information, mobile status information and the user’s behavior information.Second, train the collected cell phone data. Specially, we investigate two strategies for modeling the relevance between the stay time of mobile users and contextual information, i.e., Stay Status Prediction (SSP) and Stay Time Prediction (STP). For SSP, we use Decision Tree Algorithm and Support Vector Machine Algorithm, and Multiple Linear Regression and Principal Component Regression for STP.Third, we deploy the models trained by the two strategies on the mobile for one week testing, and then we compare and analyze the prediction results. In the analysis of the stay status prediction results, we use two baseline methods for comparison, and transform the format of STP results to fit SSP, and then analyze the six methods. In the analysis of the stay time prediction results, we mainly compare the Multiple Linear Regression method and a naive statistic method. Then we analyze the energy consumption and the memory cost of four methods.Four, we put forward a preliminary solution for Cold Start problem. There are two solutions, the first one is to build a two-dimensional mapping for user, position and stay time, and the other is to do clustering analysis for all the users.The research of Stay Time Prediction is an important complement and extension of position prediction. It will surely have a significant effect to Lo-cation Based Service, and could make the related services more accurate and more outstanding.