Penggunaan aplikasi Perplexity telah berkembang pesat, namun kajian mengenai sentimen pengguna masih terbatas. Penelitian ini bertujuan untuk menganalisis sentimen pengguna terhadap aplikasi Perplexity dengan menerapkan algoritma Random Forest. Data sebanyak 1.400 ulasan dikumpulkan melalui teknik web scraping dari Google Play Store. Selanjutnya, data melalui tahap preprocessing untuk meningkatkan kualitas teks. Untuk menangani ketidakseimbangan kelas, diterapkan teknik Random Oversampling yang menyeimbangkan distribusi kelas. Hasil penelitian menunjukkan bahwa tanpa Random Oversampling, performa model terbaik diperoleh pada N_estimator = 20 dengan akurasi 91.86%. Dengan Random Oversampling, akurasi meningkat menjadi 96.68% pada N_estimator = 30, disertai peningkatan presisi 96.76%, recall 96.68%, dan F1-score 96.68%. Konfigurasi model terbaik mencakup Max_features = 0.8, criterion = Gini, serta Max_Depth yang tidak dibatasi untuk pertumbuhan pohon yang optimal. Pengujian menggunakan K-Fold Cross Validation (K=5) memastikan kestabilan model. Secara keseluruhan, penerapan Random Oversampling terbukti efektif dalam meningkatkan akurasi dan keseimbangan model Random Forest, sehingga
menghasilkan performa klasifikasi sentimen yang lebih baik.
The use of the Perplexity application has grown rapidly, but studies on user sentiment are still limited. This study aims to analyze user sentiment towards the Perplexity application by applying the Random Forest algorithm. Data of 1,400 reviews were collected through web scraping techniques from the Google Play Store. Furthermore, the data went through a preprocessing stage to improve text
quality. To handle class synchronization, the Random Oversampling technique was applied to balance the class distribution. The results showed that without Random Oversampling, the best model performance was obtained at N_estimator = 20 with an accuracy of 91.86%. With Random Oversampling, the accuracy increased to 96.68% at N_estimator = 30, accompanied by an increase in precision of 96.76%, recall of 96.68%, and F1-score of 96.68%. The best model configuration includes Max_features = 0.8, criterion = Gini, and unlimited Max_Depth for optimal tree growth. Testing using K-Fold Cross Validation (K = 5) ensures model stability. Overall, the application of Random Oversampling proved effective in improving the accuracy and balance of the Random Forest model, resulting in better sentiment classification performance.