Classifying Application User Comments Using the Improved K-Nearest Neighbor and BM25F Weighting Methods

Abstract

Mobile JKN is an application developed by an Indonesian state-owned health insurance company, BPJS Kesehatan. The application is developed to provide easier access and more optimal service for its participants. The application allows users to access various information related to national health insurance programs anywhere, anytime. Despite the benefits this application offers, Mobile JKN received low ratings from the users. However, the ratings are sometimes irrelevant to the comments. We proposed an application to automatically review the comments based on ratings and comments, both negative and positive, using the Improved K-Nearest Neighbor (IKNN) classification algorithm with the BM25F weighing method. The initial stage of the algorithm involved document preprocessing, the document consisting of comments and rating were preprocessed to obtain the term. After that, BM25F weighing was applied to find the similarity between documents. The document was then classified using IKNN based on its BM25F weight. The test result showed that the highest accuracy rate was 0.925, obtained from parameter k=20 on IKNN with k1=1.2, Bs=0.75, and BM25F weight of 2 and 5. The result indicates that this method manages to classify the document properly