Domain adaptation of information extraction models

Abstract

Domain adaptation refers to the process of adapting an extraction model trained in one domain to another related domain with only unlabeled data. We present a brief survey of existing methods of retraining models to best exploit labeled data from a related domain. These approaches that involve expensive model retraining are not practical when a large number of new domains have to be handled in an operational setting. We describe our approach for adapting record extraction models that exploits the regularity within a domain to jointly label records without retraining any model.