A Machine Learning and Deep Learning Approach for Bengali News Headline Categorization

Abstract

Internet is a thing through which a huge amount of information and data is available. As the amount of online news is increasing drastically due to the availability of internet in all parts of the world, people are also interested in reading news from online news portals due to the availability of internet. The online news portals are- Facebook, Twitter, WhatsApp, Telegram, Instagram, Blog etc. As the amount of news is increasing in the news portals, the number of readers is also increasing. As the amount of digital data is increasing in the world, the need for data classification for that digital data is also increasing. There are several methods of data classification, such as machine learning, deep learning, etc., as well as other data mining algorithms. Data is categorized using these algorithms, so that people read the news headlines before reading the news to easily understand the main theme of the news. Natural language processing approaches are used to classify data in any language for such problems. In this research paper, Bengali news has been classified into 7 categories using machine learning and deep learning. The categories are International, National, Sports, Amusement, Politics and IT. BiLSTM, GRU, Uni-gram, Machine Learning (Logistics regression, Multinational naïve bayes, Random Forest classifier, Support vector machine) have been used to classify these categories. While the accuracy of BiLSTM is 83.42%, the accuracy of GRU is 80.01%. Among machine learning, the accuracy of Logistics regression is 64%, the accuracy of Multinational naïve bayes is 61%, the accuracy of Random Forest classifier is 65% and the accuracy of Support vector machine is 65%.