<![CDATA[Sentiment analysis is crucial in understanding public opinion, particularly in emerging technologies such as automation AI and robotic restaurant services. However, achieving accurate sentiment classification in sentiment analysis tasks poses challenges, especially when dealing with imbalanced data. This study employs the Cross-Industry Standard Process for Data Mining (CRISP-DM) through the Naive Bayes Classifier (NBC) algorithm and Synthetic Minority Over-sampling Technique (SMOTE) to address imbalanced data challenges in sentiment analysis. Social network analysis (SNA) collects and analyzes user-generated content related to automation AI and robotic restaurant services, providing insights into public sentiment. Additionally, the occurrence of frequently used words such as "people" (182), "food" (158), "jobs" (135), "robots" (137), "wage" (102), "work" (78), "robot" (79), "minimum" (78), "fast" (70), and "workers" (65) is examined. The performance of the NBC algorithm with and without SMOTE integration is compared. With SMOTE, the algorithm exhibits an accuracy of 70.11% +/- 3.52%, precision of 88.82% +/- 5.06%, recall of 46.06% +/- 6.13%, AUC of 0.967 +/- 0.016, and F-measure of 60.46% +/- 6.02%. Without SMOTE, the algorithm yields an accuracy of 48.90% +/- 4.36%, precision of 72.15% +/- 5.25%, recall of 44.32% +/- 7.15%, AUC of 0.777 +/- 0.051, and F-measure of 54.57% +/- 5.78%. Recommendations to further enhance the algorithm's performance include exploring additional optimization techniques, such as feature engineering and ensemble methods, and continuing data collection and augmentation efforts to improve dataset representativeness. Regular monitoring and evaluation and iterative refinement based on evolving data patterns are also recommended to ensure sustained effectiveness in sentiment analysis tasks.]]>
Copyrights © 2024
