Sentiment Analysis of Roblox App Reviews: Correlating User Feedback with Ratings Using Lexicon and Machine Learning Methods

The study presents a detailed sentiment analysis of user feedback for the Roblox app, focusing on correlating review sentiments with numerical ratings. Utilizing both lexicon-based and machine learning techniques, the research examined 320,000 reviews from the Google Play Store. The VADER lexicon-based analysis classified approximately 76% of reviews as positive, 18% as negative, and 6% as neutral. This distribution reflected a predominantly positive sentiment, aligning with an overall trend of user satisfaction. Pearson and Spearman correlation coefficients were calculated to evaluate the relationship between sentiment scores and user ratings, yielding moderate positive relationships with values of 0.47 and 0.36, respectively. These correlations indicated that, while sentiment scores generally paralleled user ratings, other factors likely influenced rating variations. In the machine learning analysis, the study utilized Naive Bayes and Logistic Regression models for sentiment classification. Logistic Regression achieved a slightly higher accuracy of 86.4% compared to Naive Bayes' 85.3%, showing improved precision and recall for positive and negative sentiments. However, both models struggled with the neutral category, reflecting challenges in sentiment differentiation. Cross-validation confirmed the stability of the models, with the Logistic Regression model maintaining a consistent accuracy across folds. These findings underscore the potential of sentiment analysis in providing actionable insights for developers, highlighting specific areas where user sentiment diverges from ratings. Future research could enhance sentiment detection accuracy by incorporating advanced deep learning models and extending the analysis to include data from other platforms, offering a more comprehensive understanding of user feedback across gaming ecosystems.