IMPROVING CAUSE DETECTION SYSTEMS WITH ACTIVE LEARNING

IMPROVING CAUSE DETECTION SYSTEMS WITH ACTIVE LEARNING
ISAAC PERSING AND VINCENT NG

Abstract. Active learning has been successfully applied to many natural language processing
tasks for obtaining annotated data in a cost-effective manner. We propose several extensions to an
active learner that adopts the margin-based uncertainty sampling framework. Experimental results
on a cause detection problem involving the classification of aviation safety reports demonstrate
the effectiveness of our extensions.