Penerapan SMOTE pada Algoritma Naive Bayes untuk Klasifikasi Sentimen Ulasan Wuthering Waves

Fabian Gavrilio Arya Winokan (Informatika, Fakultas Teknologi Informasi, Universitas Mercu Buana Yogyakarta)
Indah Susilawati (Informatika, Fakultas Teknologi Informasi, Universitas Mercu Buana Yogyakarta)

Jurnal Teknologi Dan Sistem Informasi Bisnis
Vol. 8 No. 3 (2026): Juli 2026

User reviews of Wuthering Waves on the Google Play Store contain important business insights. The sentiment analysis isconstrained by class imbalance and the use of gacha community slang. This condition triggers majority bias in theclassification algorithm. This research aims to build an accurate sentiment classification system to detect user complaints.The approach used is Knowledge Discovery in Databases (KDD). Pre-processing is optimized using a Gacha LexiconDictionary and Bigram extraction (TF-IDF) to resolve jargon ambiguity. The Synthetic Minority Over-sampling Technique(SMOTE) is applied to synthesize minority class data or negative reviews. The balanced data is classified using theMultinomial Naive Bayes algorithm through GridSearchCV optimization. The test results prove that SMOTE implementationsuccessfully increased the sensitivity metric (Recall) of the negative class from 0.32 to 0.82. The combination of all theseoptimizations produces a model with a final Accuracy of 86%. The integration of lexicon and SMOTE is proven to overcomedata bias. This classification model has been implemented into an interactive analytic dashboard prototype.