Identifier

etd-07082010-190449

Degree

Master of Science in Computer Science (MSCS)

Department

Computer Science

Document Type

Thesis

Abstract

Given text which is a union of d documents of strings, D = d1, d2,...., dd, the emphasis of this thesis is to provide a practical framework to retrieve the K most relevant documents for a given pattern P, which comes as a query. This cannot be done directly, as going through every occurrence of the query pattern may prove to be expensive if the number of documents that the pattern occurs in is much more than the number of documents (K) that we require. Some advanced query functionality will be required, as compared to listing the documents that the pattern occurs in, because a de_x000C_ned notion of "most relevant" must be provided. Therefore, an index needs to be built before hand on T so that the documents can be retrieved very quickly. Traditionally, inverted indexes have proven to be effective in retrieving the Top-K documents. However, inverted indexes have certain disadvantages, which can be overcome by using other data structures like suffix trees and suffix arrays. A framework was originally provided by Muthukrishnan [29] that takes advantage of the number of relevant documents being less than the occurence of the query pattern. He considered two metrics for relevance:frequency and proximity and provided a framework that took O(n log n) space. Recently, Hon et al [14] provided a framework that takes O(n) space to retrieve the Top-K documents with more optimal query times, O(P + K logK) for arbitrary score functions. In this thesis we study the practicality of this index and provide added functionalities, based on the index, to retrieve Top-K documents for specific cases like phrase searching. We also provide functionality to output the K most relevant documents(according to page rank) when two patterns are given as queries.

Date

2010

Document Availability at the Time of Submission

Release the entire work immediately for access worldwide.

Committee Chair

Shah, Rahul

DOI

10.31390/gradschool_theses.3060

Share

COinS