Efficient Approximate Entity Matching Using Jaro-Winkler Distance

Yaoshu Wang, Jianbin Qin, Wei Wang

Abstract: Jaro-Winkler distance is a measurement to measure the similarity between two strings. Since Jaro-Winkler distance performs well in matching personal and entity names, it is widely used in the areas of record linkage, entity linking, information extraction. Given a query string q, Jaro-Winkler distance similarity search finds all strings in a dataset D whose Jaro-Winkler distance similarity with q is no more than a given threshold \(\tau \). With the growth of the dataset size, to efficiently perform Jaro-Winkler distance similarity search becomes challenge problem. In this paper, we propose an index-based method that relies on a filter-and-verify framework to support efficient Jaro-Winkler distance similarity search on a large dataset. We leverage e-variants methods to build the index structure and pigeonhole principle to perform the search. The experiment results clearly demonstrate the efficiency of our methods.