Machine Learning for Record Linkage

"Integrating and querying multiple databases, web sites, or document repositories is difficult, in part because information sources often use different data formats and terminology when referring to the same entities. Machine learning offers an automatedapproach for deriving and maintaining the entity mappings necessary to link records across information sources. The approach we propose here, if successful, will significantly advance this technology. Previous research on record linkage has largelyignored the fact that each source typically has uniform conventions for representing data. By taking this into account, we can to a better job of understanding and characterizing the relationships between different sources. Specifically, it is possibleto analyze and deconstruct individual fields and thereby efficiently learn rules for mapping between records in different sources."