Clustering for data matching

Abstract

The problem of matching data has as one of its major bottlenecks the rapid deterioration in performance of time and accuracy, as the amount of data to be processed increases. One reason for this deterioration in performance is the cost incurred by data matching systems when comparing data records to determine their similarity (or dissimilarity). Approaches such as blocking and concatenation of data attributes have been used to minimize the comparison cost. In this paper, we analyse and present Keyword and Digram clustering as alternatives for enhancing the performance of data matching systems. We compare the performance of these clustering techniques in terms of potential savings in performing comparisons and their accuracy in correctly clustering similar data. Our results on a sampled London Stock Exchange listed companies database show that using the clustering techniques can lead to improved accuracy as well as time savings in data matching systems. © Springer-Verlag Berlin Heidelberg 2006.