CSCI 8970 – Colloquium Series – Fall 2010 – Fifth Event
Database Support for Private and Personalized Location-based Services
Monday, October 11, 2010
Presenter: Dr. Mohamed F. Mokbel
The fifth colloquium presentation of CSCI 8970 dealt with private and personalized location-based services. Dr. Mohamed F. Mokbel is currently working on a project that allows individuals to obtain services through the use of GPS technology without completely giving away their position or location. The Casper project does this by reducing the service-privacy trade off. While some individuals do not care about privacy and are interested only in obtaining a service (they obtain 100% service), other individuals are concerned about retaining various degrees of privacy, yet as privacy increases service tends to decrease. Casper’s location anonymizer diminishes this trade off by allowing users to switch their privacy requirements at any time. Cloaked in a wide location, an individual will receive four answers to their query, where some of which will be closer to various points within one’s cloaking area. Mokbel showed as a demonstration of this technology and how the Casper project has improved since 2007.
Answering various questions, Mokbel explained that Casper can change the user “identity” every time one submits a query. The more one moves, the more it is necessary to appear as a different person, since one increasingly separates oneself from the original group within which one was initially cloaked. They project recently expanded into possible applications for a P2P environment, a TinyCasper System (Sensornet), and it is continuously trying to improve its interface and its ability to provide services to users who wish to retain their privacy.
The second part of the colloquium dealt with the CareDB project. This project attempts to make user searching more specific. While we are currently able to find out where are the closest services, we have no way of knowing the quality of different businesses. CareDB takes those variables into account. The CareDB architecture includes user context/preference, environmental context and database context. Looking at different preference evaluation methods, the team decided to use Top-K, Skyline, K-Dom, K-Freq, and Top-K Dom in the project. The search system they designed allows the user to pick which evaluation method he/she wants to use. They then tested various approaches to make a system practical and scalable. Using a layered approached, which required only 200 lines of code, proved to provide bad or less accurate results. A built in approach provided very accurate results, but required 8000 lines of code! In comparison, the extensible approach, provided results almost as good as the built-in approach but only requiring 300 lines of code.
Currently the objectives of CareDB include providing better information through the search system as well as improving the FrexPref software prototype. Other projects Mokbel’s team is currently working on include: GeoSocialDB, FAST (A framework for Flash-Aware Search Trees), RecStore, Spatial Hadoop, Pantheon, and Predictive Queries over Spatio-temporal Streams. Attending the lecture was very informative as Dr. Mokbel is currently involved in very innovative work. One can only imagine the possibilities for his works applications. Currently Dr. Mokbel receives funding primarily from Microsoft and the NSF.
