Mobile user data mining: Mining relationship patterns

Abstract

Mobile user data mining focuses on finding useful and interesting knowledge out from raw data collected from mobile users. Frequency pattern and location dependent mobile user data mining are among the algorithm used in this field. Parallel pattern, our previous proposed method, extracts how a group of mobile users makes similar decisions, such as by moving towards the similar direction, or by viewing similar contents at the same time. Parallel pattern is triggered group behaviour of mobile users. This paper reports our refinement work on parallel pattern which incorporated refinement of the relationships among parallel patterns, or relationship pattern, which shows how 'similarities of decisions' are related to each other. Effects found are such as conditional relationship, where one parallel pattern has to happen before the next one occurs. Other effects includes associative, sequential and loop pattern effects. Our performance evaluation reports how relationship pattern performs in real life dataset and synthetic dataset and discusses some potential implementation issues. © IFIP International Federation for Information Processing 2005.