Development of a Sentiment Analysis Model for Evaluating Open Source Reviews on CCleaner

The increasing volume of user generated content on digital platforms has made automated analysis essential for understanding consumer perceptions, identifying product issues, and supporting decision making. Mobile applications in particular receive large quantities of short and informal reviews that contain valuable information about user satisfaction and expectations. This thesis develops and evaluates a sentiment analysis framework for classifying user reviews of the CCleaner Android application, with the goal of assessing model performance, identifying dominant themes, and providing data driven recommendations for product improvement. The study applies a multi model approach combining traditional machine learning algorithms, specifically Logistic Regression and Support Vector Machines, with a deep learning architecture based on Long Short-Term Memory networks. The methodology includes comprehensive preprocessing, feature extraction using TF IDF and word embeddings, model training, and evaluation with accuracy, precision, recall, and F1 score. In addition, K Means clustering is used to uncover underlying themes within user feedback and to complement the results of sentiment classification. The findings indicate that the Long Short Term Memory model outperforms traditional machine learning methods, achieving the highest overall classification accuracy and demonstrating strong capabilities for interpreting the sequential structure of short app reviews. Topic modeling results reveal recurring themes related to performance, usability, device optimization, and advertising concerns. The combined insights highlight key areas for product refinement and show the practical value of sentiment analysis for supporting the development of mobile applications. The thesis concludes with recommendations for future research and model enhancements.