An Extraction Method of Representative Patterns for QA

This paper proposes method to generate patterns which can represent documents. The proposed form of pattern consists of part-of-speech (POS) tags and surface words. The pattern can reflect linguistic information using POS tags. We make candidate patterns using N-gram. We extract top of the k patterns by proposed scoring measure using N-gram frequency and IDF. The proposed pattern extraction method can be applied to a QA system, and extract the representative patterns to an arbitrary question type.