Text Tiling Text Segmentation Based on Hierarchical Dirichlet Process Model

Abstract

Text segmentation plays an extremely important role in many areas such as abstract extraction and information retrieval. Topic model is one of important methods in text segmentation. However, the current methods based on topic model generally rely on the manual setting of the number of topics. In order to improve the efficiency of text segmentation, this paper proposes a TextTiling text segmentation method based on Hierarchical Dirichlet process (HDP) model. Firstly, the method uses the HDP model to obtain a vector representation of the text in the topic space, which can also automatically generate the number of topics. Applying the theme vector to the TextTiling segmentation algorithm can implement text segmentation. The results show that the method can effectively improve the performance of text segmentation, and be free from manually set topics.