In the digital era, public discourse on government policies has shifted significantly to online platforms. This presents valuable opportunities for governments to assess real-time public sentiment. However, prior studies on sentiment analysis in public policy remain fragmented, often lacking methodological consistency and domain-wide synthesis. This study conducts a Systematic Literature Review (SLR) to consolidate insights on the techniques, datasets, and trends involved in sentiment analysis applied to government development policies. The review identifies SVM, BERT, and Naive Bayes as the most frequently used and effective methods, with SVM excelling in structured data and simpler tasks, and BERT demonstrating superior performance in handling nuanced textual data. Lexicon based tools such as VADER are also used for quick sentiment classification. Social media platforms, particularly Twitter, emerge as the dominant data sources due to their high volume and real-time nature, while evaluation metrics such as precision, recall, F1-score, and confusion matrix are commonly applied to assess model performance. The findings also reveal evolving research interests from early focus on health policies to recent interest in infrastructure, environmental, and technology-related policies. Public sentiment across these areas varies, with health and environmental policies often eliciting negative responses, while technology policies show more neutral to positive sentiment. By synthesizing methods, datasets, evaluation strategies, and policy domains, this review provides a structured foundation to future research and supports policymakers in designing strategies.
Harwenda et al. (Fri,) studied this question.