Sentiment Analysis and Topic Modeling for Duolingo Application on Google Play Store

Knowing user satisfaction is an important thing for Duolingo to do. There are many ways that Duolingo can get feedback from users. One of them is by exploring user opinions through user reviews on Google Play. By performing sentiment analysis and topic modeling, Duolingo can find out how users view the application without having to read all the reviews that can reach hundreds of thousands. The Naive-Bayes Classifier (NBC) algorithm can be used to perform sentiment analysis. From this research, the model with a data sharing ratio of 80:20 produces excellent performance, with an accuracy of 90.51%. On the other hand, Latent Dirichlet Allocation (LDA) was used to perform topic modeling. The results of this study found that the optimum number of topics widely discussed in Duolingo user reviews is 5 topics, each with positive and negative sentiments. Based on this study, most Duolingo users leave reviews with positive sentiments. Most users felt that Duolingo is a good app and extremely helpful in learning foreign languages. However, a few users also expressed their complaints about the Duolingo application in terms of features, application performance, and content. This research shows that Naive-Bayes Classifier (NBC) and Latent Dirichlet Allocation (LDA) are very suitable and able to work well in sentiment analysis and topic modeling on user reviews of the Duolingo application.