Date of Award

11-28-2016

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Information Science

First Advisor

John Talburt

Abstract

This dissertation is focused on the methodology of determining the quality of the results of entity resolution. Entity resolution methodologies results in different success rates. Until now these success rates have been measured by counting all of the correctly and incorrectly matched results. This is quickly becoming an intractable task as datasets grow in size. In order to address this problem this research tests and describes the results of count based measures as compared with inferred measures borrowed from the information retrieval community. The key contribution of this research is a proof of concept in a controlled environment that demonstrates that sampling without a known truth can work for assessing ER error rates. This work introduces a bridge between the research communities of IR and ER by completing a proof of concept to demonstrate a novel method for ER error rate estimation. Experimental validation is then used to establish that a method for ER error rate estimation using sampling is possible and that a truth or benchmark set is not required for evaluation of ER error rates. This research is novel and significant because in previous research the commonly accepted method of Entity resolution evaluation has been count based measures.

Download

Share

COinS