As mobile devices become location-aware, they offer the promise of powerful new applications, such as location- enhanced instant messaging [10] and digital graffiti sys- tems [5, 6]. However, fulfilling these promises requires overcoming a number of tough challenges. Our research concen- trates on one of these problems, the acquisition and use of personal gazetteers. 数据挖掘研究院
A personal gazetteer records places that are meaningful for a specific person. While different applications may re- quire different information about places, for us the essential requirements are a textual label and some sort of geomet- ric representation, e.g., a point, a set of points, or a region. Thus, my personal gazetteer might include places such as: • Espresso Expose - latitude: 44o5702’, longitude: 93o1548’. • Home - lattitude: 445698’, longitude: 93o1527’. 数据挖掘研究院
We have argued elsewhere [11] for the advantages of per- sonal gazetteers based on a notion of place, rather than just physical location. For example, there may be a com- plex relationship between what a person considers a place and physical location: “Target” might refer to any one of a chain of discount stores within a metropolitan area. Further, the descriptions people use to refer to places – for example, “the coffeeshop”, “Espresso Expose”, or “the off-brand cof- fee place next to Big 10” – vary depending on who the de- scriptions are produced for and the purpose for which they are produced. More generally, research in environmental psychology explores how people naturally structure their ex- perience around personally and socially meaningful places - home, office, school, church, coffeeshop, pub, etc. [8] [9] [12] 数据挖掘实验室
Once we commit to building systems around users’ per- sonal places, we face a new problem: how is a system to acquire personal gazetteers? This paper takes an interac- tive discovery approach to this problem. Specifically, we developed a clustering algorithm that discovers a user’s per- sonal gazetteer from the user’s spatiotemporal histories, i.e., time-stamped location data. We then embed this algorithm in an interactive system that lets users visualize and confirm, modify, or reject the places discovered by the system.
