A Reliable Technique for Sentiment Analysis on Tweets via Machine Learning and BERT

In recent times, when we fiddle with innovation particularly with social media, it leads to an increasing change in human conduct. The day-by-day utilization of web-based media by individuals has expanded a lot that it is gradually infusing an expression into our conduct. Particularly, Twitter is generally utilized authoritatively by numerous individuals and is one of only a handful few social media platforms where individuals search for reliable information. The intent of this work is to perform sentiment analysis of the tweets using different Machine Learning algorithms and the BERT model to prevent offensive use of words and give an insight into the emotion of the words in the tweet. The dataset utilized is obtained from Kaggle. The Machine Learning models employed are Support Vector Machine (SVM), Stochastic Gradient Descent (SGD), Logistic Regression, Decision-Tree., Random Forest Method, and XGBoost along with a deep learning architecture BERT for the analysis. A comparative analysis of all the models was carried out and it can be concluded that logistic Regression obtained the highest training and validation accuracy of 82.65% and 81.74%, respectively, and the F1 score of this model was 79.14%. BERT model has an exceptional performance in Natural Language Processing, so BERT was also implemented for the sentimental analysis of the Twitter tweets and it achieved the highest training accuracy, validation accuracy, and F1 score of 94%., 93.63%, and 92.34% respectively.