In this paper we present and validate a novel approach for single-label multi-class document categorization. The proposed categorization approach relies on the statistical property of Principal Component Analysis (PCA), which minimizes the reconstruction error of the training documents used to compute a low-rank category transformation matrix. This matrix allows projecting the original training documents from a given category to a new low-rank space and then optimally reconstructs them to the original space with a minimum loss of information.The proposed method, called Minimum Loss of Reconstruction Information (mLRI) classifier, uses this property, extends and applies it to unseen documents. Several experiments on three well-known multiclass datasets for text categorization are conducted in order to highlight the stable and generally better performance of the proposed approach in comparison with other popular categorization methods.
展开▼
