Sentiment Analysis on User Reviews of Mutual Fund Applications

Abstract

The primary goal of this study is to compare the accuracyof the results of sentiment analysis using the Naive Bayes, Support VectorMachine (SVM), and Random Forest methods on one of the mutual fund application’s user reviews.The second goal is to identify user reviews of the mutual fund app to gaininsight into the topics covered by each sentiment. The user reviews have beencollected through a web scraping method on the google play store, then cleanedthrough several processes of data pre-processing. Feature extraction wasperformed using TF-IDF along with vectorization using n-grams. The modelperformance was measured using a confusion matrix. Using a ratio of 80:20 ontraining and testing data, resulting in an accuracy of 92.7, 93.7 and 94.2% forNaive Bayes, SVM, and Random Forest methods, respectively. Identify the topicscovered by each sentiment in user reviews using visualizations. In the positivesentiment of users, the majority discusses the application which is easy andgood, especially for novice investors. In negative sentiment, the majoritydiscussed the slow sales process to disbursement of funds and long loadingtimes when opening the application.