User Profile in Absence of Ground Truth for Mobile Users

Mobile user personalization is used to increase user engagement on many platforms. One way to achieve personalization is by building user profiles, which encompass certain attributes of mobile phone users. Mobile manufactures and service providers often collect data of their users in order to provide personalized services. This collected data is rich in user behavior, but seldom has enough ground truth information collected directly from users to build a user profile. In this paper, we address this problem and provide a framework for developing user profile of mobile users in absence of ground truth data. Our approach consists of a one-class classification technique to address this issue. We test our method on data of one million mobile phone users and show that the predicted accuracy is close to that achieved using a supervised model. We further extend this method to predict other attributes of the user, again getting a good accuracy.