An improved method of statistical model for text segmentation

Abstract

Every document contains multiple topics, and the task of text segmentation is to segment a text into several parts that each part represents one topic. On the base of statistical model of Masao Utiyama whose experiment showed that the method was more accurate than or at least as accurate as a state-of-art text segmentation system, this paper proposes an improvement suggestion trying to improve the existing problem. The experiment results showed that the improved algorithm improved both the efficiency and the accuracy.