Topic Detection with Hypergraph Partition Algorithm

An algorithm named SMHP (Similarity Matrix based Hypergraph Partition) algorithm is proposed, which aims at improving the efficiency of Topic Detection. In SMHP, a T-MI-TFIDF model is designed by introducing Mutual Information (MI) and enhancing the weight of terms in the title. Then Vector Space Model (VSM) is constructed according to terms' weight, and the dimension is reduced by combining H-TOPN and Principle Component Analysis (PCA). Then topics are grouped based on SMHP. Experiment results show the proposed methods are more suitable for clustering topics. SMHP with novel approaches can effectively solve the relationship of multiple stories problem and improve the accuracy of cluster results.