Demo Link --> https://youtu.be/pnN0sStTL7k?feature=shared
Sentiment analysis is essential in the social media age for comprehending user behaviour and public opinion. This study looks into how well machine learning models work for sentiment anal- ysis of COVID-19-related Twitter data. The study investigates the effectiveness of classifiers like Naive Bayes (NB), Random Forest (RF), and Support Vector Machine (SVM) in classifying tweets into positive, negative, and neutral sentiments using a dataset of tweets manually labeled with sentiment categories and evaluates and ensemble of the architecture too. We also implement Bidirectional Encoder Representations from Transformer (BERT). The models’ decisions are in- terpreted, and insights into the features underlying sentiment classification are obtained using the Explainable Artificial Intelligence (XAI) approach - Shapley Additive Explanations (SHAP). The outcomes show how well these models work at reliably identifying the emotion expressed in tweets connected to COVID-19, providing insight into the general public’s sentiment patterns during the pandemic and helping decision-makers across various industries. For RF, NB, SVM, and BERT, we obtain an accuracy of 96%, 91%, 94%, and 80.21%, respectively