SVM-Based Sentiment Analysis for Bug Tracking Priority Scale

Abstract

The rapid expansion of mobile applications in Indonesia creates a massive density of user reviews, making manual categorization of technical bugs highly inefficient for development teams. This research develops an automated bug tracking system powered by sentiment analysis to address these constraints. Utilizing a dataset of 1,200 reviews extracted from the Google Play Store (2023-2024), the study compares the Support Vector Machine (SVM) algorithm with a Radial Basis Function (RBF) kernel against the Naive Bayes classifier. The methodology encompasses automated scraping, text preprocessing via the Sastrawi stemming library, and TF-IDF feature weighting to handle semantic ambiguities in informal Indonesian text. Evaluation results demonstrate that SVM achieves a superior accuracy of 86.25% and an F1-score of 0.86, significantly outperforming Naive Bayes which stands at 77.50%. McNemar’s test yields a p-value of 0.006, statistically validating SVM’s effectiveness in reducing False Negatives. While highly successful in rule-based priority scaling, the architecture automatically classified the actionable technical bug reports into 150 Critical, 200 Major, and 50 Minor urgency categories. Operationally, the automated system processes data 99.8% faster than manual methods, cutting down the identification period from 12 hours to just 85 seconds. These findings imply that embedding SVM into bug tracking architectures facilitates real-time mitigation of critical stability issues. While highly successful, future studies should focus on overcoming linguistic limitations regarding sarcasm detection through advanced deep sequential processing.