Klasifikasi Ulasan Pengguna Aplikasi: Studi Kasus Aplikasi Ipusnas Perpustakaan Nasional Republik Indonesia (PNRI)

Abstract

The decline in the number of new user trends for the iPusnas application affects the achievement of the target value of the LKIP Pujasintara PNRI report 2020-2024. This is related to the rating value of application user reviews on Google Playstore which is still considered lower than other similar applications. Electronic Word of Mouth (EWOM) which is very influential on the decision of prospective new users of the application in considering the best similar application, because it involves reviewing the rating value and user reviews. Several previous studies have proven that difficulties are always encountered when analyzing or extracting important information in user reviews of applications manually. Review analysis is very useful for developing application service features in order to increase user satisfaction and application value ratings, so it requires an automatic user review classification tool by finding the best suitable model. This study applies the CRISP-DM methodology, but only until the evaluation stage. The classification algorithm used is Naïve Bayes, Logistic Regression, Support Vector Machine (SVM), and a combination of tf-idf unigram, bigram, and trigram features. The result of this research is that the combination of the tf-idf unigram (F1) feature with the SVM algorithm achieves the best value for each evaluation value of precision, recall, and f1-score of 87% each. The lowest evaluation value is precision 55% from the combination of F2 features with SVM, 42% recall and 32% f1-score from the combination of F3 features with logistic regression.