On the analysis of users' behavior based on mobile phone apps

Abstract

Currently, we live in highly connected environments in which we consume large amounts of data, such as contextual information, and produce even larger amounts of data. In this never-ending cycle, the generated data is taken to other applications, creating pervasive and context-aware systems with which we interact. As this cycle goes on, users share their personal and private data, such as current location, activities, and even their mood, which establish a drawback as users may not want to expose themselves. Therefore, we face a trade-off between sharing personal data and, thus, losing privacy, or not sharing it and, possibly, losing quality of experience in relation to services and applications. In this work, we analyze properties of a mobile phone dataset containing precise information about users' accesses to applications to answer the following questions: (i) What can be inferred from an user at the current time given his/her past information? (ii) How does location data, which is a fundamental information in these networks, affect the inferences? To do this, we apply supervised and unsupervised learning techniques to predict network type - Mobile or WiFi, application name, and user ID. Our results present an overview of how knowledge can be extracted from data shared by users, and which type of data is the most revealing.