Perbandingan Kinerja Multinomial Naïve Bayes pada Variasi Rasio Data Training dan Testing dalam Analisis Sentimen Ulasan Taman Rakyat Slawi Ayu

Abstract

Sentiment analysis is one of the text mining techniques used to identify public opinions toward a particular object or service. Taman Rakyat Slawi Ayu, as one of the public green spaces in Tegal Regency, receives numerous visitor reviews through Google Maps, which can be utilized as a source of information to evaluate the quality of its services and facilities. This study aims to compare the performance of the Multinomial Naïve Bayes algorithm using different training and testing data split ratios for sentiment analysis of Taman Rakyat Slawi Ayu reviews. The data were collected through a web crawling technique on Google Maps, resulting in 4,153 reviews. After the data selection process, 1,794 reviews met the analysis criteria. The preprocessing stage consisted of cleaning, case folding, tokenizing, stopword removal, and stemming, resulting in 1,789 reviews ready for classification. Sentiment labeling was performed based on user ratings, where ratings of 1–3 were categorized as negative sentiment and ratings of 4–5 as positive sentiment. The experiments were conducted using three data split ratios: 70:30, 80:20, and 90:10. The results showed that the 90:10 ratio achieved the highest accuracy of 87.71%, with a precision of 0.88, recall of 0.99, and F1-score of 0.93. However, this ratio used a smaller testing dataset, making the evaluation results less representative than those obtained using the other data split ratios. Overall, the study demonstrates that variations in the training and testing data split ratio affect the classification performance of the Multinomial Naïve Bayes algorithm in sentiment analysis of Taman Rakyat Slawi Ayu reviews.