An Experimental Study to Investigate the Use of Additional Classifiers to Improve Information Extraction Accuracy

In this paper, we present an information extraction system and investigate the use of additional classifiers to help improve information extraction performance. We propose a simple idea of training an additional classifier using the same feature configurations on another corpus and then using this new classifier to classify the original dataset. The classification result of this new classifier is then used as a feature to the original classifier. We tested this approach on the CMU seminar announcements and the Austin job posting datasets and obtained results better than all previously reported systems.