Search Efficiency in Indexing Structures for Similarity Searching.pdfVIP

Search Efficiency in Indexing Structures for Similarity Searching Girish Motwani Sandhya G Department of Computer Science and Automation Indian Institute of Science 4 0 0 November 24, 2003 2 r a Abstract M Various indexing structures have been proposed 2 Similarity searching finds application in a wide vari- for similarity searching in metric spaces. We present 1 ety of domains including multilingual databases, com- the performance analysis of these structures in terms putational biology, pattern recognition and text re- of the percentage of database scanned by varying edit ] trieval. Similarity is measured in terms of a distance distances from 10% to 100%. B function (edit distance) in general metric spaces, which After providing a preliminary background in Section D. is expensive to compute. Indexing techniques can be 2, we move on to the description of the existing s used reduce the number of distance computations. We index structures in Section 3. Section 4 describes the c present an analysis of various existing similarity index- experimental set up and the analysis is presented in [ ing structures for the same. The performance obtained Section 5. Section 6 concludes the paper. 2 using the index structures studied was found to be un- v satisfactory . We propose an indexing technique that 4 1 combines the features of clustering with M tree(MTB) 2 Preliminaries 0 and the results ind