摘要：An algorithm named SMHP (Similarity Matrixbased Hypergraph Partition) algorithm is proposed, whichaims at improving the efficiency of Topic Detection. InSMHP, a T-MI-TFIDF model is designed by introducingMutual Information (MI) and enhancing the weight of termsin the title. Then Vector Space Model (VSM) is constructedaccording to terms' weight, and the dimension is reduced bycombining H-TOPN and Principle Component Analysis(PCA). Then topics are grouped based on SMHP.Experiment results show the proposed methods are moresuitable for clustering topics. SMHP with novel approachescan effectively solve the relationship of multiple storiesproblem and improve the accuracy of cluster results.
