Prediction of Bengali Newspaper Headline Polarity Using Machine Learning Techniques

Abstract

Automated detection of the sentiment of newspaper headlines can save time and effort in reading and help to create a positive environment by reducing the prevalence of negative news and promoting positive news. Negative news can have a variety of effects on society, including Increased fear and anxiety, Desensitization to violence and tragedy, Decreased trust in institutions, Polarization, and division, and Decreased well-being. In this study, we propose a system for detecting the positivity or negativity of newspaper headlines in Bangla. While there has been a significant amount of research on detecting the sentiment of English newspaper headlines, there has been none in Bangla. To build our dataset, we used a large number of Bangla newspaper headlines. To categorize these Bangla newspaper headlines, three classic machine learning models, e.g. support vector machine (SVM), logistic regression, and random forest, and one deep learning model, e.g. sequential model are utilized. We then applied these three classical machine learning models as well as a deep learning model to the dataset following a machine learning approach. The obtained results are quantified in terms of four inquisitive performance metrics – accuracy, F1-score, precision, and recall. The results showed that the sequential deep learning model surpassed all other classical machine learning models achieving an accuracy of 80% F1-Score of 79% Precision of 87% and Recall of 76%.