Sentiment analysis on Bangladesh airlines review data using machine learning

Abstract

A common means of transportation in our everyday lives is air travel. As a result, it's no surprise that more and more customers are posting their airline reviews online. However, in the age of machine learning, it would be much easier to extract millions of pieces of information and knowledge from them if a model was used to polarize and comprehend them. Sentiment analysis may be used to understand people's attitudes or sentiments by utilizing sites that provide opinion-rich data. In this work, we worked on a customized dataset including online reviews for 4 major Bangladesh Airlines, performed a multiclass sentiment analysis, and compared the classi ers. Alongside sentiment analysis, topic modeling is also done to get better decisions based on the actual experiences of other customers who have own with airlines. This method begins with pre-processing procedures used to clean the reviews and balance the review data using the Pegasus model's oversampling mechanism. The analysis was carried out 3 di erent machine learning (Decision Tree, Random Forest, and XGBoost) and 3 di erent deep learning classi cation strategies (CNN, LSTM, BERT). The test set's output is the review sentiment (positive/negative/mixed) using a three-class dataset, and the performance in terms of accuracy is calculated. Based on the results, we have achieved the best accuracy 83% in terms of BERT. The accuracies were determined to compare each categorization technique, and the total sentiment count for all four airlines of Bangladesh was displayed in terms of domestic route, international route and overall route. We comprehend the results acquired from USA airlines Tweets data and demonstrate that our framework is more e cient than the earlier model. Therefore, it is essential to consider whether a sentiment makes a particular prediction. Thus, we then train an interpretable LIME model for the sentiments and the construction of explainable sentiments can have a major advantage.