Text Segmentation Model Based on Multiple Discriminant Analysis
Jingbo Zhu · Journal of Software · 2007
This paper proposes a new domain-independent statistical model. In this model, four multiple discriminant analysis (MDA) criterion functions are defined and used to achieve global optimization in finding the best segmentation by means of the smallest within-segment distance, the largest between-segment distance and segment length. To alleviate the high computational complexity problem introduced by the new model, genetic algorithms (GAs) are used. Comparative experimental results show that the methods based on MDA criterion functions have achieved higher Pµ than that of TextTiling and Dotplotting algorithms.