Learning Name Variants from Inexact High-Confidence Matches

摘要: Name variants which differ more than a few characters can seriously hamper record linkage. A method is described by of first names and surnames be learned automatically from records that contain information needed for true link decision. Post-processing limited manual intervention (active learning) unavoidable, however, to differentiate errors in the original digitised data variants. The demonstrated on basis an analysis 14.8 million Dutch vital registration.