Access Type

Open Access Dissertation

Date of Award


Degree Type


Degree Name



Computer Science

First Advisor

Farshad Fotouhi


Research on similarity join techniques is becoming one of the growing practical areas for study, especially with the increasing E-availability of vast amounts of digital data from more and more source systems. This research is focused on pre-processing clustering-based techniques to improve existing similarity join approaches.

Identifying and extracting the same real-world entities from different data sources is still a big challenge and a significant task in the digital information era. Dissimilar extracts may indeed represent the same real-world entity because of inconsistent values and naming conventions, incorrect or missing data values, or incomplete information. Therefore discovering efficient and accurate approaches to determine the similarity of data objects or values is of theoretical as well as practical significance.

Semantic problems are raised even on the concept of similarity regarding its usage and foundation. Existing similarity join approaches often have a very specific view of similarity measures and pre-defined predicates that represent a narrow focus on the context of similarity for a given scenario. The predicates have been assumed to be a group of clustering [MSW 72] related attributes on the join. To identify those entities for data integration purposes requires a broader view of similarity; for instance a number of generic similarity measures are useful in a given data integration systems.

This study focused on string similarity join, namely based on the Levenshtein or edit distance and Q-gram. Proposed effective and efficient pre-processing clustering-based techniques were the focus of this study to identify clustering related predicates based on either attribute value or data value that improve existing similarity join techniques in enterprise data integration scenarios.