QuIET: A Text Classification Technique Using Automatically Generated Span Queries

Abstract

We propose a novel algorithm, QuIET, for binary classification of texts. The method automatically generates a set of span queries from a set of annotated documents and uses the query set to categorize unlabeled texts. QuIET generates models that are human understandable. We describe the method and evaluate it empirically against Support Vector Machines, demonstrating a comparable performance for a known curated dataset and a superior performance for some categories of noisy local businesses data. We also describe an active learning approach that is applicable to QuIET and can boost its performance.