SBIR Phase II: A Machine Learning Approach to Approximate Record Matching

Choicemaker Technologies, Inc., New York NY

Investigators

Abstract

This Small Business Innovation Research (SBIR) Phase II project will enhance the company's approximate record-matching software, the Maximum Entropy De-Duper, MEDD(TM) by: 1) Enhancing MEDD's performance using advanced standardization tools to convert data, such as names and addresses, into standard formats; 2) Expanding MEDD's market by matching business names not only person names; 3) Internationalizing MEDD to support Canadian French or Mexican Spanish; 4) Benchmarking MEDD against the competition and developing a methodology to objectively compare matching systems; 5) Reducing MEDD's reliance on training data to ease deployment; producing the best possible "untrained" models that will adapt and improve through client use; 6) Applying the latest advances in machine learning technology to the record-matching problem to increase competitive advantage; and 7) Speeding MEDD word blocking with a fast, innovative memory-resident data-store. MEDD's market includes all business and government entities that store mission-critical information in large databases. The project will yield societal benefits for public health, anti-terrorist efforts, epidemiological research, the U.S. Census, and the data quality of records relating to racial and ethnic minorities.