Evaluating Public Satisfaction with Public Service Applications using Supervised Machine Learning

Public service applications play a pivotal role in ensuring seamless interaction between citizens and government services. However, evaluating the effectiveness of these applications and understanding public satisfaction remains a challenge. This study focuses on developing a robust supervised machine learning framework to assess public satisfaction by analyzing user feedback collected from application reviews. By leveraging advanced text preprocessing techniques, feature extraction methods such as Term Frequency-Inverse Document Frequency (TF-IDF) and word embeddings, and implementing classification algorithms like Support Vector Machines (SVM) and Random Forests, this framework provides a comprehensive analysis of satisfaction levels. The dataset, consisting of over 10,000 reviews categorized into positive, neutral, and negative sentiments, was subjected to rigorous preprocessing and balanced to ensure model fairness. Evaluation metrics, including accuracy, precision, recall, and F1-score, were used to benchmark the models, with SVM demonstrating superior performance with an accuracy of 89.5%. The findings reveal key determinants of satisfaction, highlighting the importance of sentiment polarity and textual features. This research provides valuable insights for developers and policymakers to enhance the user experience and adapt public service applications to meet evolving citizen expectations. Future work aims to incorporate hybrid models and extend this methodology to other public service domains to further improve satisfaction analysis accuracy and utility.