Text Classification Using Sparse Representation Classifier

Abstract

Text classification or categorization is the process of assigning structured or unstructured newlinetext documents to predefined categories or labels. The text can be offline or newlineonline and of any size. It is gaining greater attention with the surge in the volume newlineof online data available. Text classification is an integral part of the analysis of newlinetext data with wide-ranging applications like document retrieval, opinion mining, newlineemail classification, and spam filtering. The text classification process consists of newlinestages like data acquisition, data analysis and labeling, feature construction, feature newlineweighting, feature selection, feature projection, and classifier design. This thesis newlineproposes methods to apply the idea of sparse representation in designing the classifier newlinefor text classification. newlineFrom the literature survey done, it is observed that classical algorithms like k- newlineNearest Neighbour (k-NN), Support Vector Machine (SVM), Naive Bayes, Neural newlineNetworks (NN) and their variations yielded only reasonable results in addressing the newlineproblem, leaving enough room for further improvement. A class of computationally newlinecomplex algorithms referred to as Deep Neural Networks (DNNs) were also used newlinefor text classification and attained better accuracy than the classical methods. But it newlinedemands a large volume of training data and high computational facilities, and the newlinemodel lacks interpretability. A very few works are there based on the idea of Sparse newlineRepresentation. A class of algorithms commonly referred to as sparse methods has newlineemerged recently from compressive sensing and found numerous applications in newlinemany areas of data analysis and image processing. Sparse methods as a tool for newlinetext analysis is an alley that is largely unexplored rigorously. Based on widespread newlineand successful usage of sparse-based representation methods in various applications newlinesuch as bioinformatics, image denoising, deblurring, restoration, image retrieval, and face recognition.