Improving Naive Bayes Sentiment Classification on Spotify Reviews Through Three-Body Problem Optimization

Digital platforms today depend on user satisfaction, making sentiment analysis a key tool for business insight. This study explores how Spotify user reviews can be classified using the Naive Bayes (NB) model, chosen for its clear and interpretable structure. While NB is simple and transparent, its default setup often falls short in terms of accuracy. To improve performance, the smoothing parameter alpha is fine-tuned using the Three-Body Problem Optimization Algorithm (3BPOA), a newer metaheuristic inspired by gravitational interactions. The results show a jump in accuracy from 91.83% to 93.33%, along with improved F1-scores, especially for classes that typically perform poorly. Beyond metrics, the optimized model produces meaningful keyword lists for each sentiment class, which can inform product and marketing decisions. These top words reflect real user concerns and preferences, offering practical value to business teams. Although optimization adds some computational cost, the gains in performance and clarity justify the trade-off. Lightweight models like NB, when paired with targeted tuning, can meet business demands for both accuracy and interpretability. This approach supports smarter decision-making without sacrificing transparency, bridging the gap between model performance and practical application in real-world platforms.