Aspect-based Sentiment Analysis of Mobile Apps Reviews using Class Association Rules and LDA

As the number of people using smartphones grow, mobile apps have become increasingly important in our lives. Users find it difficult to evaluate the apps. Software developers also has need to get feedback from users to determine how to enhance the app’s usability by recognizing drawbacks and strengths. We can conclude that sentiment analysis of App reviews comes to overcome these challenges. Because users express their opinions about different aspects of the app, feature-based sentiment analysis is now an effective strategy. As a next phase for sentiment classification, many studies have focused on adopting machine learning or natural language processing. Despite its strengths, it has some disadvantages. One of its disadvantages is that it fails to account for the inwardness of the reviews. For that reason, class association rules become a good choice to identify language patterns of reviews and classifying them based on these patterns. This research applies Latent Dirichlet Allocation (LDA) for feature extraction and class association rule Apriori algorithm for aspect-based sentiment classification. Our approach achieved accuracy improvement by 5.68% and 7.58% when it is applied on a dataset of Angry Birds game app reviews which outperform traditional machine learning naive bayes and SVM algorithm.