A machine learning and deep learning approach for Baengali newspaper headline categorization

Abstract

The internet world is called a repository of information and data. Where there is a huge amount of information and data collection. Through internet people can access any kind of information and data from any place at any time. Current technology has made information and data readily available, due to which the amount of online news on the Internet has increased tremendously. Furthermore, because the internet is so widely available, people are growing increasingly eager to read news articles from news websites that use direct data. In general, online news portals are the terms used to describe Facebook, Twitter, WhatsApp, Telegram, Instagram, blogs, and other services. The quantity of news available on internet news portals is growing daily, and this growth is being matched by an increase in readers. All this online news are digital data, and with the volume of digital data is growing, so is the requirement for data categorization. Numerous methods, including machine learning, deep learning, transfer learning, and other data mining techniques, may be used to classify data. These algorithms classify data such that readers may deduce the news story's primary idea from the headlines alone. To address such issues, data in any language may be classified using natural language processing techniques. This article divides Bengali news stories into six categories: Politics, entertainment, sports, national, international, and IT. It does this by using deep learning and machine learning techniques. Numerous techniques, including BiLSTM, GRU, and Uni-Gram, as well as conventional machine learning algorithms, including SVM, MNB, RF Classifier, and LR, are used to select these classifications. The accuracy rates for these models are as follows: GRU achieves 84.01% accuracy, BiLSTM attains 83.42% accuracy, Logistic Regression performs at 64%, Multinomial Naive Bayes scores 61%, Random Forest Classifier achieves 65% accuracy, and Support Vector Machine also achieves 65% accuracy.