Employing Lexicalized Dependency Paths for Active Learning of Relation Extraction

Active learning methods which present selected examples from the corpus for annotation provide more efficient of supervised relation extraction models, but they leave developer in unenviable role a passive informant. To restore developer’s proper as partner with system, we must give an ability to inspect model during development. We propose make this possible through representation based on lexicalized dependency paths (LDPs) coupled active learner LDPs. apply LDPs both simulated and real ACE evaluation year’s newswire training show that greatly reduces cost while maintaining similar performance level learning, yields comparable state-of-the-art using small number annotations.