Sentiment analysis on airline reviews using Naive Bayes / Nurul Sarah Aliessa Che Harun

Airlines get a vast number of customer feedback each day but manually analysing them may be time-consuming and incorrect. The unstructured framework of these reviews, along with class inequalities, make it difficult to extract useful information. This study used the Naive Bayes algorithm to analyse airline reviews and customer feedback, resulting in improved service quality. This may include a preliminary design and implementation evaluation that is further divided into three phases. At the preliminary phase, the literature review was performed, and there was a selection of the dataset composed of 9,210 Airline Reviews from 2020 to 2023 from Kaggle. In the design and implementation phases, TF-IDF was considered in cleaning and preparing the reviews for application, while the construction of a sentiment analysis model was implemented through the use of the Multinomial Naive Bayes algorithm. A friendly GUI was developed for the ease of reviews analysis. Three splits were taken for testing of the model, out of which 70:30 produced the best results, having AUC (Area Under the Curve) to be 0.89, whereas the accuracy is 84.58%. In order to tackle class imbalance in the dataset, ADASYN oversampling was done and then, afterward, the model turned to more reliable regarding detecting positive and negative sentiments. The findings proved that this algorithm of Naive Bayes is appropriate and efficient to analyze reviews provided for airlines. This would really give the opportunity to understand customer feedback clearly, along with faster problems fixation, resulting in an enhanced service quality being developed. This technique in the near future can further be enhanced through complex algorithms by applying it on bigger datasets also, which creates the platform for enhancement in customers' satisfaction and makes airlines smarter too.