An Interactive Search Technique for String Databases

The explosive growth of string databases makes similarity search a challenging problem. Current search tools are non-interactive in the sense that the user has to wait a long time until the entire database is inspected. We consider the problem of interactive searching, and propose a set of innovative -NN ( -Nearest Neighbor) search algorithms. We propose a new model for the distance distribution of a query to a set of strings. Using this distribution, our first technique orders the MBRs (Minimum Bounding Rectangles) of an index structure based on their order statistics. We also propose an early pruning strategy to reduce the total search time for this technique. Our second technique exploits existing statistical models to define an order on the index structure MBRs. We also propose a method to compute the confidence levels for the partial results. Our experiments show that our technique can achieve 75 accuracy within the first 2.5-35 of the iterations and 90 accuracy within the first 12-45 of the iterations. Furthermore, the reported confidence levels reflect the quality of the partial results accurately.