Date of Award

Fall 2011

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Computing Sciences - (Ph.D.)

Department

Computer Science

First Advisor

James Geller

Second Advisor

Narain Gehani

Third Advisor

Dimitri Theodoratos

Fourth Advisor

Michael Halper

Fifth Advisor

Soon Ae Chun

Sixth Advisor

Hayato Yamana

Abstract

The search terms that a user passes to a search engine are often ambiguous, referring to homonyms. The results in these cases are a mixture of links to documents that contain different meanings of the search terms. Current search engines provide suggested query completions in a dropdown list. However, such lists are not well organized, mixing completions for different meanings. In addition, the suggested search phrases are not discriminating enough. Moreover, current search engines often return an unexpected number of results. Zero hits are naturally undesirable, while too many hits are likely to be overwhelming and of low precision.

This dissertation work aims at providing a better Web search experience for the users by addressing the above described problems.To improve the search for homonyms, suggested completions are well organized and visually separated. In addition, this approach supports the use of negative terms to disambiguate the suggested completions in the list. The dissertation presents an algorithm to generate the suggested search completion terms using an ontology and new ways of displaying homonymous search results. These algorithms have been implemented in the Ontology-Supported Web Search (OSWS) System for "famous people."

This dissertation presents a method for dynamically building the necessary ontology of "famous people" based on mining the suggested completions of a search engine. This is combined with data from DBpedia. To enhance the OSWS ontology, Facebook is used as a secondary data source. Information from "people public pages" is mined and Facebook attributes are cleaned up and mapped to the OSWS ontology.

To control the size of the result sets returned by the search engines, this dissertation demonstrates a query rewriting method for generating alternative query strings and implements a model for predicting the number of search engine hits for each alternative query string, based on the English language frequencies of the words in the search terms. Evaluation experiments of the hit count prediction model are presented for three major search engines. The dissertation also discusses and quantifies how far the Google, Yahoo! and Bing search engines diverge from monotonic behavior, considering negative and positive search terms separately.

Share

COinS