Modeling of mobile end-user context

Abstract

Emerging mobile services have spawned new revenue sources, such as messaging, Internet browsing and multimedia. One of the business opportunities that mobile industry has not yet fully exploited is the contextual status of end-users. Context-aware systems are gaining importance in telecommunications since the applications are numerous and have relevance from the industrial (e.g. in aspects involving user segmentation) and academic (e.g. analysis of mobile service adoption dynamics) points of view.This thesis first presents a theoretical discussion: evolution of telecommunications services, prior studies in context-aware systems and other concepts such as data mining techniques and network theory, and network visualization. The second part focuses on the development of a context detection algorithm. This algorithm extracts contextual information from data logs containing cell-id transitions. It follows two steps in context detection: first a clustering process where physically close cells are grouped into clusters and second the context detection for every one of those clusters by using time-based assumptions. The thesis uses a handset-based tool in collecting data logs. The strength and accuracy of the algorithm are tested through analysis of the output files. Finally, a study of real data (from the Finnish market) is carried out in order to deliver results. Through this analysis, the thesis focuses on the service usage perspective. The driving question is how and where the end-users spend their time with the handsets.Context is not only about location but also about the physical status and social settings of end-users. Context detection provides a new dimension for example in service usage analysis or modeling of service adoption. The results show that e.g. most of the usage of WLAN takes place at "home" and applications such as "Navigation and Maps" or "Browsing" are used "on the move" context. Intensity graphs prove that e.g. "Home" is not the most active context despite being the most frequent one and the intensity of usage abroad is most active in "Multimedia" and "Messaging" applications. On the other hand, there is a significant business opportunity in applications that automatically identify the context of mobile users (all their movements along the day). Targeted marketing and handset-based contextual adaptation are some of the examples of possible applications. At last, the thesis confirms that network visualization tools are useful in the process of context modeling, not only for testing the results but also to help in the detection by using all their functionalities (as e.g. clustering). Some examples using one specific tool will be provided at the end.