Extracting User Profiles From Mobile Data

As smartphones have been increasingly adopted by mobile phone users, the applications market for mobile operating systems also developed into a giant market for advertising business. Smartphone companies allow third parties to develop apps to provide different services to users and third party app developers publish these applications in app markets relevant to the mobile operating system. Users download from a range of app categories including games, productivity, education, social networking, etc. Processing smartphone data is very appealing, since app usage on a mobile phone indicates important clues about the lifestyle and interests of the user. Although processing the data collected from smartphones is very attractive, usually the data is not available to the research community. Usually governmental organizations impose regulations on the telecommunications companies as far as the disclosure of the data collected by these companies. The data to be serviced publicly has to be organized such that it is anonymized and the identity of the customers cannot be tracked. In this paper, we propose a probabilistic model to predict the user profile of a customer, namely gender and age, based on usage behaviors of mobile applications, phone brand and model. The probabilistic model infers these demographics from the statistical characteristics profiled from a huge pool of information hidden in a dataset provided by TalkingData company. Building the probabilistic model on this dataset is a more practicable user profiling method, since it can be used to deliver personalized services without posing privacy risks to the users and it eliminates the necessity for continuously tracking users' online activities or smartphone usage and maintaining historical records. Our experimental results indicate that promising information can be tracked down through the proposed approach, especially information about user tendencies which are of crucial importance to the targeted advertisement.