News Classification Using Local Text Pattern and TF-IDF Features

In this modern era, news documents are available in large volumes in online news portals. Based on news contents they can be categorized into different groups e.g., politics, sports, etc. Thus, the automatic classification of news text can provide news of desired categories to the users. In this paper, we propose an approach that aims to categorize a given news text document with high accuracy. Our approach consists of four modules pre-processing, feature Extraction, signature Generation, and classification. Firstly, a news document is pre-processed using a rule-based approach to remove less important words. Term Frequency-Inverse Document Frequency (TF-IDF) is used for extracting statistical features. Then, we extend the concept of the Local Binary Pattern of image processing for the first time in the text classification domain for extracting the Local Text Pattern feature. We combine these features to generate a signature for each document. Finally, the Support Vector Machine is fine-tuned for training and testing. Our approach uses the BBC news dataset for experiments and obtains superior performance in terms of accuracy of 99.63% over the state-of-the-art approaches.