Department of Computer Engineering
S E M I N A R
C-TREE: Searching with Ad-Hoc Weighted Composite Metrics
Mehmet Can Kurt
Computer Engineering Department
Multimedia databases often rely on a form of similarity searching rather than exact matching. One approach to similarity search is to model database objects in a metric space defined by a distance function. In many applications, search quality can be significantly enhanced by employing multiple similarity measures, each capturing a complementary perspective. The relative importance of the measures can change over time, or users may want to emphasize certain measures that best capture their view of similarity. It is essential to support a flexible ranking strategy for each query based on a weighted combination of available metrics. We propose the C-tree and the C-forest index structures that organize objects in a way to enable ad-hoc weighting of metrics at query time. We compare our methods with the M3-tree, an extension of the M-tree to multiple metrics and demonstrate significant improvements.
DATE: 15 March, 2010, Monday @ 16:40
PLACE: EA 409