PhD dissertation defence

 If you are interested in database research in our department, you may want to attend the following:

PhD Dissertation Defense

Xi Zhang

224 Bell Hall
Friday, April 16, 2010, 10am

Committee
Professor Jan Chomicki (Chair)
Professor Hung Q. Ngo
Professor Michalis Petropoulos


Title: Probabilities and Sets in Preference Querying

Abstract:

User preferences in databases are attracting increasing
interests with the boom of information systems and the trend of
personalization. In the literature, there are two different frameworks
dealing with this topic, namely quantitative approaches and qualitative
approaches. The former assume the availability of a scoring function,
while the latter do not. Instead, in qualitative approaches,
preferences are expressed using preference formulas. We investigate
three advanced topics on preferences stemming from those frameworks.

First, we study top-k queries over uncertain data in the quantitative
framework. We formulate three intuitive semantic properties for top-k
queries in probabilistic databases, and propose Global-Topk query
semantics which satisfies them to a great degree. We also design
efficient dynamic programming algorithms for query evaluation.

Second, we observe that so far all work on top-k queries in
probabilistic database focus on ordinal scores, while there are
applications where cardinal scores are more appropriate. This
motivates our work on preference strength, where we consider the
magnitude of score in addition to the order it establishes over
tuples.

Finally, as a counterpart to the top-k query in the quantitative
framework, we explore the set preference problem in the qualitative
framework. Observing the fact that preferences can also be collective
in this case, our goal is to tackle this second-order problem with
first-order tools. We propose a logical framework for set
preferences. Candidate sets are represented using profiles consisting
of scalar features. This reduces set preferences to tuple preferences
over set profiles.  We also propose algorithms to compute the "best"
sets effectively.