Performance Improvement of Sentiment Analysis of Customer Reviews on Delivery Services Using a Naïve Bayes Model Optimised Naïve Bayes Model with PSO

Abstract

This research is motivated by the rapid growth of the delivery service industry and the importance of customer feedback in competition, especially for ID Express which has a mobile application. The main issue raised is how to analyse the sentiment of customer reviews on the ID Express app on the Google Play Store to improve service quality. User reviews, although rich in information, have not been optimally utilised, making it difficult for companies to understand user perceptions. In this study, we develop a new method to analyse the sentiment of ID Express app user reviews. This method integrates Naïve Bayes algorithm with Particle Swarm Optimisation (PSO)-based feature selection optimisation technique to produce more accurate analysis. The method used includes collecting user review data from the Google Play Store (2020-2023), preprocessing the data, implementing the Naïve Bayes algorithm, and applying PSO for feature selection. Model performance was tested with accuracy and F-measure metrics using 90:10 and 80:20 data sharing ratios. The results showed that the Naïve Bayes algorithm with PSO produced 52% accuracy at 90:10 ratio and 63% at 80:20 ratio, with F-measure values of 43% and 55% respectively. In conclusion, the use of PSO as feature selection improves the accuracy and F-measure of sentiment analysis using Naïve Bayes, especially at a data sharing ratio of 80:20.