Study on Sub Topic Clustering of Multi-Documents Based on Semi-Supervised Learning

Sub-topic detecting is an important step in the abstracting of multi-documents.This paper describes a new method for sub-topic detecting based on semi-supervised learning:it firstly gets the primal sets of topics by hierarchy clustering,and labels the sentences which have high scores in the topics,then use the method of constrained-kMeans to decide the number of topics(k),and finally get the topic sets by k-Means clustering.The experiment result indicates that its value is stable.