<![CDATA[Online reviews have become a primary source for consumers in evaluating the quality of products and services, including in Bali’s culinary sector. However, these reviews are typically written in informal language, containing slang, non-standard expressions, and code-mixing between Indonesian, English, and local languages. Such linguistic characteristics pose challenges for sentiment analysis systems, as language variability can reduce the consistency and reliability of classification results. This study compares the performance of two machine learning methods, namely Naïve Bayes and Support Vector Machine (SVM), in classifying sentiment from unstructured Balinese culinary reviews. The dataset consists of 5,000 Google Reviews, which were processed through text cleaning, slang normalization, and TF-IDF feature representation using a combination of unigram and bigram models. Performance was evaluated using accuracy, precision, recall, F1-score, and 5-fold cross-validation to assess model stability. Naïve Bayes was employed as a baseline to enable a more objective comparison between the two methods. The experimental results indicate that SVM achieves more balanced and consistent performance than Naïve Bayes, particularly for minority classes. Naïve Bayes tends to be biased toward the majority class due to its independence assumption and the imbalanced data distribution. Furthermore, the preprocessing stage, especially slang normalization, contributes to reducing lexical variability and improving the interpretability of sentiment patterns. These findings suggest that appropriate classifier selection and tailored preprocessing strategies play a crucial role in maintaining the reliability of sentiment analysis systems in informal and multilingual contexts.]]>
Copyrights © 2026
