A Supervised Learning Model for Sentiment Analysis Based on Regional Dialects in Tourism-Related Issues

Abstrak

Indonesia has an exceptionally rich diversity of regional languages, one of which is the Bekasi dialect, often used in social media communication. The uniqueness of this dialect presents specific challenges in extracting public opinion, especially in text-based sentiment analysis. This study aims to develop a sentiment analysis framework that incorporates regional dialects from social media data and evaluate the effectiveness of various supervised learning algorithms. Data were collected from the Facebook group “Explore Bekasi Tourism,” totaling 1,257 posts and comments, which were filtered down to 1,000 relevant instances. A manual validation process was conducted by linguistic experts to convert non-standard terms and regional dialects into standardized Indonesian, followed by translation into English for annotation purposes. The analysis method involved preprocessing steps (tokenizing, case folding, stemming), feature weighting using TF-IDF, and sentiment classification using four algorithms: Naive Bayes, K-Nearest Neighbor, Support Vector Machine, and Decision Tree. The evaluation results show that Naive Bayes achieved the best performance with an accuracy of 76%, followed by K-Nearest Neighbor (67.5%), SVM (65.5%), and Decision Tree (28%). These findings highlight the crucial role of expert judgment in processing dialect-based data to ensure accurate sentiment classification. The study recommends developing a broader annotated corpus of regional dialects and exploring deep learning methods in future research to enhance classification performance and generalizability.