This paper addresses the domain adaptation problem in natural language processing (NLP) from an instance weighting perspective. The authors formally analyze the domain adaptation problem, identifying two distinct needs for adaptation: differences in instance distributions and classification functions between the source and target domains. They propose a general instance weighting framework that integrates several adaptation heuristics into a unified objective function. The framework aims to remove misleading training instances, assign higher weights to labeled target instances, and augment training instances with predicted labels from target instances. Empirical results on three NLP tasks—part-of-speech (POS) tagging, named entity (NE) recognition, and spam filtering—show that incorporating more information from the target domain through instance weighting significantly improves performance compared to regular semi-supervised and supervised learning methods. The paper also discusses related work and highlights the importance of accurately setting instance weighting parameters.This paper addresses the domain adaptation problem in natural language processing (NLP) from an instance weighting perspective. The authors formally analyze the domain adaptation problem, identifying two distinct needs for adaptation: differences in instance distributions and classification functions between the source and target domains. They propose a general instance weighting framework that integrates several adaptation heuristics into a unified objective function. The framework aims to remove misleading training instances, assign higher weights to labeled target instances, and augment training instances with predicted labels from target instances. Empirical results on three NLP tasks—part-of-speech (POS) tagging, named entity (NE) recognition, and spam filtering—show that incorporating more information from the target domain through instance weighting significantly improves performance compared to regular semi-supervised and supervised learning methods. The paper also discusses related work and highlights the importance of accurately setting instance weighting parameters.