This study presents a comparative analysis of Support Vector Machine (SVM) and Bidirectional Encoder Representations from Transformers (BERT) for sentiment analysis on electric motorcycles in Indonesia using data from the social media platform X, formerly known as Twitter. The dataset of 128,711 tweets collected between 2015 and 2024 was refined through systematic preprocessing, reducing the corpus to 38,954 entries after data cleaning, tokenization, and feature selection. The objective was to evaluate algorithm performance in classifying public sentiment, with metrics including accuracy, precision, recall, and computational efficiency. Results showed that SVM achieved higher overall accuracy 89.74% with strong precision for positive sentiment 91%, while BERT, specifically the IndoBERT variant, demonstrated superior recall for negative sentiment 91% despite slightly lower accuracy 87.90%, effectively capturing nuanced contextual language, such as sarcasm, informal expressions, and emotionally ambiguous statements that require deeper semantic understanding beyond literal word meanings. Computational analysis revealed that SVM required approximately 53 minutes of CPU training, compared to BERT’s 3.3 hours on GPU. The study suggests that SVM is optimal for rapid, resource-constrained applications, whereas BERT excels in detailed contextual analysis. These findings guide stakeholders in selecting algorithms based on analytical priorities, such as monitoring public reception or addressing consumer concerns
Adiwinata et al. (Mon,) studied this question.