Polarity Prediction of News Headlines of Bengali Newspapers using Machine Learning

Abstract

Time is becoming more and more scarce resource keeping pace with the advancement of technology. Automatic detection of polarity of newspaper heading can save time of reading. It helps to analyze social status to reduce the number of bad news. It provides a platform for serving good news and create a positive environment. In this project work, we develop a system that detects the positivity or negativity of a newspaper headline in Bangla. It has been studied a lot earlier, but all of the research work has been done in English based on English Newspapers. It is a matter of fact that no work has been done in Bangla. We use a large number of Bangla newspaper's headlines in order to build our data set in this project work since there was no such work before in Bangla. Web crawler is used to create the Bangla data set. Then we use three classifiers based on machine learning, e.g. support vector machine (SVM), logistic regression and random forest and one classifier based on deep learning, e.g. sequential model. We thoroughly investigate the feasibility and applicability of the classifiers. We train and test the data set with them following machine learning approach. Sequential deep learning model outperforms all other classifier achieving an accuracy of 80%, SVM shows the poorest result, i.e. an accuracy of 55%.