ELM: An Extended Logic Matching Method on Record Linkage Analysis of Disparate Databases for Profiling Data Mining

摘要

As predictive marketing and customer profiling solutions have become more sophisticated, they have increasingly become dependent on data from external sources. In order to utilize this data, records must be linked to internal records without the use of unique identifiers. The Extendable Logic for Matching (ELM) performs probabilistic matching from disparate sources and classifies matches according to discrete values reflective of their utility. Sets of matching rules are evaluated based on their performance on supervised classification tasks. High performance on a classification task is indicative of congruity with the real-world entity concerned, giving a sense of matching quality without the use of a gold standard. A set of matching rules generated using name and address was compared to a set which was matched using exact string comparison. We conclude that exact string comparison is a superior method for matching on highly sparse demographic data from disparate sources.