A Difference Latent Semantic Indexing
Yibin Song · Journal of Yantai University · 2008
On the base of analysis of global LSI and local LSI,a new difference latent semantic indexing is proposed,which integrates the class information into term set.Medical web pages are used to test the new LSI.The text in medical webpage is extracted and represented by the global LSI and the difference LSI respectively.SVD and SLSI are used to reduce the dimension of feature space,SVM algorithm is employed to classify the feature vectors of testing collection,and the categorical accuracy and macro-average F1 are calculated.Experiment illustrates that the difference LSI gives higher accuracy and macro-average F1 than the global LSI when combined with SVD or SLSI.However,the difference LSI combines with SLSI can't obtain more improvement on accuracy and the macro-average F1.