Mining Spatiotemporal Data: From Personal Use to Community Knowledge

点击翻译按钮获取中文摘要

This project will develop data mining algorithms for extracting concepts such as semantic locations and routines from location based service traces, along with techniques for handling privacy concerns. The traces will come from a system called PlaceMail that generates use records that link the virtual and physical worlds. PlaceMail users create virtual messages (such as reminders and to-do lists) for delivery at a specific place or time on a location-aware cell phone client. PlaceMail also can collect traces of user presence at a location and time. These data will enable novel applications such as a local search engine that lets the user tap into the collective community opinion to find the best place to obtain desired goods or services, and a social matching system that can bring together people with similar routines, to facilitate ride sharing and increase community cohesion.The algorithms will extract and process information from location-linked data, including data mining algorithms to extract higher-level constructs like paths from traces and retrieval algorithms that combine keywords from messages with popularity estimates based on traces to return relevant and high-quality results. Also to be developed will be social matching algorithms that can find people with similar routines and recommendation algorithms to suggest new places a user might be interested in.Privacy concerns are a central issue for such a system, so an important project of the project will be designing easy-to-use techniques for managing them. Privacy requirements will be investigated through contextual prototyping with PlaceMail users, in which early prototypes of the applications are filled in with the users' own data. This method will make it easy for users to conceive of the risks and benefits of sharing their data, and thus will lead to reliable requirements. Empirical evaluations of the applications will validate results. The algorithms and privacy management techniques will be incorporated into target applications. Large field trials of PlaceMail (to generate sufficient amounts of real data) will be combined with controlled laboratory studies that compare the target applications to existing baseline applications. Location-aware technologies can link the new virtual world of the World Wide Web to the physical world, and this project is a major step in the development of tools that can aggregate useful location-linked information. Not only will people benefit individually, for example by finding better information about places in their community, but also communities may benefit from increased cohesion, as citizens learn from each other's experience, establish relationships based on shared interests, and use public resources (roads, community centers, parks, and the like) more efficiently. This project will share the datasets produced with the research community, suitably anonymized and with user consent, and it will provide deep and diverse training to a set of graduate students.