Approximate Address Matching

Address management is a major challenge for many organizations, as errors occur frequently in the address capturing process, and address standards and usages may vary among different databases. Rather than comparing house number, street, city and province individually, we use a string similarity measurement to perform address comparison, which enables us to combine the edit distance with the vector space model to search for potentially matching address candidates by associating them with a similarity matching score. Upon evaluating the strengths and weaknesses of these techniques, we introduce an algorithm for effective address matching, called Term-Weighted Dissimilarity, which combines edit distance similarity with Term Frequency-Inverse Document Frequency weighting. We implement this algorithm in software and show its effectiveness via a real application for address matching and correction based on Canada Post´s address standard.