Precision Driven Sentiment and Topic Classification of News Articles using IndicBert

This paper explores the application of machine learning models for sentiment analysis and topic classification of news articles. We employed multiple models including Logistic Regression, Multinomial Naive Bayes, and advanced transformer-based models like IndicBERT to classify news articles based on their sentiment and topics. The study utilized a dataset of news headlines from Indian Express, leveraging TF- IDF vectorization and BERT embeddings for feature extraction. Our results show that Logistic Regression on IndicBERT embeddings outperforms traditional methods in terms of accuracy, precision, and recall. We also demonstrate the effectiveness of these models through confusion matrices and performance metrics, highlighting the potential for improving news classification systems. The Logistic Regression model achieved strong performance with an accuracy of 87.5%, precision of 65%, recall of 70%, and an F1 score of 68%, showing good balance in sentiment classification. Similarly, the Multinomial Naive Bayes model performed comparably with an accuracy of 87.1%, precision of 61%, recall of 68%, and an F1 score of 65%.