Aspect-Based Sentiment Analysis of User Reviews on the Game “Honkai: Star Rail” Using Naïve Bayes Classifier

Abstract

Game is a form of entertainment that is often used to refresh the mind from the fatigue of daily activities and routines. Honkai: Star Rail is a popular turn-based game from Hoyoverse available on Google Play Store. Several studies have proposed Sentiment Analysis with Naïve Bayes classification method. However, not many have identified the reviews of a game to the extent of identifying on its aspects. In aspect-based sentiment analysis, text is analyzed to identify various attributes or components, then the relevant sentiment (positive, negative, or neutral) for each of these attributes is determined. This research aims to analyze aspect-based sentiment using the Naïve Bayes Classifier method, as well as categorize sentiment into positive and negative, and classify reviews into certain aspects. The results obtained after 5-fold iteration obtained the best average accuracy of 79%, The evaluation results show that it is necessary to tune the model using Grid Search Hyperparameter Tuning. Optimization of smoothing parameters with alpha = 0.1 proved effective in improving model performance with the highest weighted average accuracy of 93%. The evaluation results show that Grid Search Hyperparameter Tuning optimization gives better performance to the Naive Bayes algorithm model in multi-label classification.