The notion of preference occurs naturally in every context where one talks about human decision or choice. This project studies preferences in the context of database queries. Faced with information overload, database users seek ways to obtain not necessarily all answers to queries but rather the best, most preferred answers.
In the framework of this project, user preferences are captured as preference formulas. Preference queries involve the use of a number of algebraic preference operators that have simple formal semantics. The most basic of those is winnow which, when applied to a relation, returns the set of the most preferred tuples -- those which are not dominated by any other tuples in the relation. The winnow operator is parameterized by a preference formula. For example, given a suitable formula and a database of books for sale winnow will return all the cheapest ways to purchase every book. Other preference operators include ranking (unbounded iteration of winnow) and preference range selection.
In the course of the project algorithms for evaluating preference operators are developed and studied. Algebraic properties of those operators are identified, in order to lay the foundation for the optimization of preference queries. Query optimization techniques for such queries are also developed and integrated with an existing query optimizer. The research addresses all the steps necessary to make preference queries a practical concept in the area of database management.
Further information can be obtained at the project web site (www.cse.buffalo.edu/~chomicki/prefs.html). Broader impacts include new courses for students, new applications to make databases better, and foundation and directions for new research.
