In this thesis, we use PDF textbook as data resource, focus on comparing the conceptual sentences of two domain-specific terms .We first calculate the mutual information of every word in sentence and selected feature words to build MI vector space model. The vector space model is used to evaluate the similarity of two sentences for the hierarchical clustering algorithm. After clustering, we choose representative labels and comparative sentence pair for every cluster. According representative labels, the clusters which have the same labels will be grouped as a new concept hierarchy.