News Categorization Using Least Squares Twin Support Vector Machine

The accelerating increase in online news content has led to more efficient methods for classifying content since traditional Support Vector Machine (SVM) methods are hindered by limitations in processing and scalability. In this study, an original implementation of Least Squares Twin Support Vector Machines (LSTWSVM) for classifying news articles is developed to create an indepth framework classifying content within eight categories of Sports, Science, Technology, Entertainment, Health, Business, National, and World news. Our choice of LSTWSVM was motivated by its proven strengths in computational efficiency and ability to handle large-scale datasets, addressing the performance limitations of conventional SVM implementations directly. The approach employs advanced natural language processing (NLP) methods, including stemming, tokenization, and stop-word elimination, in addition to TF -IDF feature extraction to maximize content representation. Empirical testing with a het-erogeneous corpus of web news stories confirmed LSTWSVM's performance superiority with classification accuracy higher than 85% and computation overhead lowered by 30% compared to traditional SVM methods. These remarkable enhancements in both efficiency and accuracy validate LSTWSVM's value in real-world content management systems, providing more accurate classification abilities and better information retrieval capabilities that ultimately translate into better user experiences in online news sites.