Latent Factor SVM for Text Categorization

Text categorization is an important research in nature language process and content analysis. In this paper, we present latent factor SVM (LF-SVM) for text categorization which use latent factor vectors for category representation on text categorization. We prove that latent factors extracted by PLSA (probability latent semantic analysis) can span convex structure to express text category. Based on the category expression we adopt maximal margin hyper plane to divide the categories. The experiments on normal text datasets show that our motivation and algorithm are reasonable and effective.