Discovering personal gazetteers: An interactive clustering approach

Personal gazetteers record individuals' most important places, such as home, work, grocery store, etc. Using personal gazetteers in location-aware applications offers additional functionality and improves the user experience. However, systems then need some way to acquire them. This paper explores the use of novel semi-automatic techniques to discover gazetteers from users' travel patterns (time-stamped location data). There has been previous work on this problem, e.g., using ad hoc algorithms [13] or K-Means clustering [4]; however, both approaches have shortcomings. This paper explores a deterministic, density-based clustering algorithm that also uses temporal techniques to reduce the number of uninteresting places that are discovered. We introduce a general framework for evaluating personal gazetteer discovery algorithms and use it to demonstrate the advantages of our algorithm over previous approaches.