Evaluasi Kinerja Naive Bayes dan K-Nearest Neighbors pada Analisis Sentimen Ulasan Aplikasi SnackVideo

Abstract

Abstrak  

Penelitian ini membandingkan performa algoritma Naive Bayes (NB) dan K-Nearest Neighbors (KNN) dalam analisis sentimen review aplikasi SnackVideo. Dataset berisi 2000 ulasan dengan distribusi seimbang antara sentimen positif dan negatif. Metodologi penelitian mengikuti CRISP-DM, meliputi pengumpulan data melalui web scraping, preprocessing data (transform case, tokenization, stopword removal, dan stemming/lemmatization), pembentukan model klasifikasi dengan NB dan KNN, evaluasi model menggunakan metrik akurasi, recall, presisi, dan AUC, serta komparasi hasil. Hasil evaluasi menunjukkan NB memiliki akurasi 80,46%, recall 88,26%, presisi 77,71%, dan AUC 96,90%, sedangkan KNN memiliki akurasi 78,84%, recall 75,82%, presisi 81,56%, dan AUC 92,70%. Secara umum, NB lebih unggul dalam akurasi, recall, dan AUC, sehingga direkomendasikan sebagai algoritma yang lebih efektif dan efisien untuk analisis sentimen pada dataset besar. KNN lebih cocok untuk kasus yang membutuhkan presisi tinggi dalam prediksi positif.

Kata kunci: Analisis Sentimen, Aplikasi SnackVideo, Naive Bayes, K-Nearest Neighbors, Preprocessing Data, Evaluasi Model

Abstract

This study compares the performance of the Naive Bayes (NB) and K-Nearest Neighbors (KNN) algorithms in sentiment analysis of SnackVideo app reviews. The dataset consists of 2000 reviews with a balanced distribution between positive and negative sentiments. The research methodology follows CRISP-DM, including data collection via web scraping, data preprocessing (transform case, tokenization, stopword removal, and stemming/lemmatization), sentiment classification model building with NB and KNN, model evaluation using metrics such as accuracy, recall, precision, and AUC, and comparison of results. The evaluation results show that NB achieves an accuracy of 80.46%, recall of 88.26%, precision of 77.71%, and AUC of 96.90%, while KNN achieves an accuracy of 78.84%, recall of 75.82%, precision of 81.56%, and AUC of 92.70%. Overall, NB outperforms KNN in terms of accuracy, recall, and AUC, making it a more effective and efficient algorithm for sentiment analysis on large datasets. KNN is more suitable for cases requiring high precision in positive predictions.

Keywords: Sentiment Analysis, SnackVideo App, Naive Bayes, K-Nearest Neighbors, Data Preprocessing, Model Evaluation