Publication

Abstract : This study focuses on the crucial task of news category classification using a dataset consisting of 175,995 instances spanning across 30 distinct categories. The system utilizes a preprocessing pipeline to collect intricate linguistic features. Pre-trained word embeddings such as GloVe, FastText and Word2Vec are utilized to gather semantic information. The dataset is balanced using SMOTE analysis, and semantic information is gathered using pre-trained word embeddings. The study investigates various machine learning algorithms such as K-Nearest Neighbors, Random Forest, XGBoost, Decision Tree, Logistic Regression, Gaussian Naive Bayes, and AdaBoost. The primary emphasis lies in attaining exceptional precision, F1 score, and comprehending the interpretability of the models’ predictions. The study also highlights the need of model interpretability, utilizing LIME. The framework for real-world news classification employs a variety of embeddings, with KNN standing out as particularly effective. It improves precision and efficiency, especially when combined with Word2Vec embeddings, demonstrating machine learning’s potential in a wide range of news categories.