Towards characterizing and exploiting fine-grained user behaviors in mobile system

Abstract

Globally, number of mobile devices and connections reached 8 billion recently, and is projected to surpass 1.5 devices per capita by 2021. These mobile devices are generating an enormous amount of data, associated with behaviours of the users and their associated context. However, most of these subtle behaviours aren’t being utilized in current mobile systems, due to difficulties associated with exploiting them. For example, some key characteristics of user behaviours such as mobility, content interest and device interaction vary both spatially and temporally, as well as being context-dependent. The challenges not only lie in finding the optimal approach of understanding these fine-grained dynamics of user behaviours, but also integrating the insights into the system. This thesis presents three main contributions that aim to improve mobile system performance, provide improved services and user experience, by exploiting aspects of fine-grained human behaviour when users are “on the move”, namely users’ interest in content (content interest), their interaction with mobile devices and their mobility. It demonstrates the feasibility of leveraging these behaviour through collecting data at various vantage points in a mobile system (i.e. cellular/wifi network, content servers and user devices). It presents mechanisms to apply findings obtained through the analysis of the data collected to develop new system architectures, more efficient applications and algorithms to improve the performance of the overall system. Firstly, we exploit commuter content consumption behaviour, as well as network path diversity to offer significant benefits to the user both in terms improved latency, and reduced cost. Secondly, by characterizing users’ behaviour when interacting with mobile devices, we show how to increase battery life and thus the user experience. Finally, with the analysis of user mobility, we show that by combining mobility with fine-grained geo-social contexts, it is possible to significantly improve the efficiency of mobile content distribution and provide better assistance for disaster management. The methods and algorithms proposed are evaluated through analytical modelling, data-driven simulations and experiments, and show that it is possible to use the information about human behaviour, that can be collected at the various vantage points of a mobile system, to significantly improve mobile system efficiency, and user quality of experience.