Development and Deployment of Sentiment Analysis AI on Citizens’ Feedback in Goiás

Abstract

This study presents the development and deployment of a sentiment analysis model based on Natural Language Processing (NLP) techniques in the context of public service evaluation. The model was implemented to automate the classification of citizens’ feedback on services provided by the EXPRESSO program in the State of Goiás, Brazil. Before implementation, the classification process was manual, which was time-consuming, and prone to inconsistencies. The dataset used included a wide range of citizen experiences, ensuring that the model captured a comprehensive and representative sample of comments. Preprocessing techniques such as noise removal, tokenization, stemming, and customized stopword adjustments were applied to refine the data for analysis. Among the algorithms tested, Multinomial Naïve Bayes (MNB) stood out for its slightly superior performance in identifying negative feedback — a critical class for monitoring service quality and addressing citizen concerns efficiently. The model was validated with managers to ensure its practical application and was subsequently integrated into the State's Data Warehouse (DW) to feed a real-time monitoring dashboard. Additionally, a visual interface was developed to allow managers to analyze feedback on demand. This interface includes features such as word clouds for positive and negative feedback, facilitating the quick identification of key themes, trends, and recurring issues. The integration of artificial intelligence into the EXPRESSO Program establishes a scalable and replicable framework to improve decision-making processes, enhance operational efficiency, and promote inclusivity. This methodology can be extended to other public service centers, emphasizing the vital role of AI in fostering responsive governance, improving service quality, and enhancing citizens’ satisfaction.