ABSTRACT
A semantic predication network was developed by processing the documents on schizophrenia into se-mantic predication sets using SemRep, from which the core information was extracted to produce a graphic summary which was consisted of highly cohesive cliques.The automatic methods for summarizing biomedical documents were studied using the network properties combined with semantic information.The subthemes in the summary obtained by clustering were evaluated according to the clique co-node matrix and the contents of the summary were assessed according to the reference criteria.The accuracy was 0.93, the recall was 0.68, and the F-value was 0.79 for the summary, indicating that this method can effectively recognize the core information in documents and the semantic information in network graphics.