Sentiment Analysis of Bangladeshi Digital Newspaper by Using Machine Learning and Natural Language Processing

The modern era has witnessed a heightened level of global news exposure in recent years. We all are updated with any news, be it the economic news of our country, the political news, national or international calamities, disease outbreaks, accidents, or homicides through various means of communication. Despite the advancements in news media, the enduring appeal and sustained demand for newspapers have not waned over time. A notable segment of the population still relies on the regular distribution of newspapers to their homes as a means of staying updated on domestic and global events. The profound influence of print media on the community and its perception of the nation is a matter of great significance. Within the realm of news, it is evident that the presence of positive and negative news yields distinct effects on individuals. This study focuses on conducting sentiment analysis of news articles featured on the front pages of prominent newspapers in Bangladesh. The polarity of news stories from five distinct newspapers was examined individually, as well as the collective polarity of the full dataset. The overall dataset consists of positive, negative, and neutral news, accounting for 54.305
^%
, 42.434
^%
, and 3.259% respectively. The classification has been done on a unique customized dataset containing the news for six consecutive months on the front pages of the leading five newspapers and the analysis was performed using six classification models: Logistic Regression, Multinomial Naive Bayes, Support Vector Machine (SVM), Support Vector Classifier (SVC), Random Forest, and K Nearest Neighbors. Overall, these models achieved accuracies of approximately 85.19
^%
, 77.39
^%
, 83.17%, 84.23
^%
, 80.60%, and 58.78% respectively.