The Application of Naive Bayes in Analyzing Public Sentiment Toward the Performance of the North Sumatra Regional Government in Handling Flash Floods

Figure Ilustration AI

FORMOSA NEWS - Medan - AI Analysis Reveals Public Dissatisfaction With North Sumatra Flood Response. A new study from researchers at Prima Indonesia University found that public sentiment toward the North Sumatra Regional Government’s handling of flash floods was overwhelmingly negative, with 88.4 percent of online opinions criticizing emergency response and aid distribution efforts. The research, published in 2026 in the Formosa Journal of Computer and Information Science, used artificial intelligence and machine learning to analyze public reactions across social media and online news platforms. The study was conducted by Rivaldo Siburian, Rikki Josua Tampubolon, Valentino Surbakti, M. Irvandy Haris, and Rizky Rahmansyah from the Informatics Engineering Program, Faculty of Science and Technology, Prima Indonesia University. The findings highlight how digital public opinion can serve as a real-time indicator of government performance during natural disasters. Researchers say the results demonstrate the growing importance of artificial intelligence in evaluating public trust, crisis management, and disaster-response policies in Indonesia.

Social Media Became a Public Evaluation Platform
Flash floods remain one of Indonesia’s most frequent hydrometeorological disasters, particularly in North Sumatra Province. During emergencies, citizens increasingly turn to platforms such as X, Facebook, and online news comment sections to express criticism, frustration, appreciation, and demands for accountability. The research team analyzed 1,132 validated public opinion entries collected between November 2025 and February 2026 through web crawling techniques and third-party APIs. Initial data collection exceeded 1,500 entries, but duplicate posts and irrelevant comments were removed to improve data quality. Researchers focused on public discussions related to flood management, evacuation speed, logistics distribution, and government communication. Queries such as “North Sumatra flood,” “Medan flood,” and “government flood response” were used to gather relevant discussions from digital platforms. According to the study, public dissatisfaction surged during the emergency-response phase of the disaster, especially when evacuation efforts were perceived as slow and poorly coordinated.

Artificial Intelligence Used to Measure Public Sentiment
The research applied sentiment analysis, a branch of natural language processing that classifies opinions into positive, negative, or neutral categories. The team primarily used the Multinomial Naive Bayes algorithm, a machine-learning model commonly used for large-scale text classification. Before analysis, the text data underwent several preprocessing stages, including text cleaning, tokenization, stopword removal, and stemming. Researchers then transformed the cleaned text into numerical representations using the TF-IDF method so that machine-learning algorithms could interpret the information. To improve labeling consistency, the researchers used the Indonesian Sentiment Lexicon (InSet) dictionary and additional human annotator validation.
The results showed a highly unbalanced sentiment distribution:

• Negative sentiment: 88.4 percent (1,001 entries).
• Positive sentiment: 9.3 percent (105 entries).
• Neutral sentiment: 2.3 percent (26 entries).

The study identified two major areas receiving the strongest criticism:

• Emergency response and evacuation speed.
• Distribution of aid and logistics.

Public communication from local officials also received negative evaluations, particularly regarding the clarity and consistency of official information during the crisis.

SVM Achieved the Best AI Performance
The researchers compared four machine-learning models to determine which algorithm performed best for sentiment classification:

• Support Vector Machine (SVM).
• Multinomial Naive Bayes.
• Logistic Regression.
• Random Forest.

Among all models, SVM achieved the strongest overall performance with an F1-Score of 0.855 and accuracy of 89.87 percent. Multinomial Naive Bayes and Logistic Regression each recorded 89.43 percent accuracy with an F1-Score of 0.844, while Random Forest produced an accuracy score of 87.22 percent. Although SVM delivered slightly better classification results, the researchers emphasized that Multinomial Naive Bayes remains highly effective because of its computational efficiency and faster processing speed for large-scale public opinion data. The research also tracked how sentiment changed throughout different disaster phases. During the early warning period, sentiment was mostly neutral because discussions centered on factual weather information. Negative sentiment sharply increased during the emergency-response stage in January 2026, when citizens discussed panic, infrastructure damage, and delayed assistance. Positive sentiment began appearing more frequently during the recovery phase in February 2026, especially in discussions praising volunteers and humanitarian aid distribution.

Implications for Government Policy and Crisis Management

The findings suggest that AI-based sentiment analysis can function as an objective digital auditing tool for evaluating government disaster-response performance. According to the Universitas Prima Indonesia research team, public opinion data collected from digital platforms can help governments identify weaknesses in emergency response systems and improve communication strategies during crises. The researchers wrote that sentiment analysis can provide “an objective instrument for government performance auditing,” especially during rapidly evolving emergency situations. The study also recommends that local governments develop real-time sentiment monitoring dashboards to detect public concerns faster and respond more effectively during disasters. Researchers further suggested that future studies should explore advanced artificial intelligence models such as IndoBERT and Transformer architectures to improve the accuracy of Indonesian-language sentiment analysis. Techniques such as SMOTE were also recommended to address class imbalance issues in highly negative datasets.

Author Profiles
Rivaldo Siburian is an Informatics Engineering researcher at Universitas Prima Indonesia specializing in machine learning and sentiment analysis.

Rikki Josua Tampubolon focuses on data mining and natural language processing research.

Valentino Surbakti studies artificial intelligence applications for digital public opinion analysis.

M. Irvandy Haris specializes in information systems and machine-learning technologies.

Rizky Rahmansyah focuses on text mining and digital policy evaluation research.

Sources
Rivaldo Siburian, Rikki Josua Tampubolon, Valentino Surbakti, M. Irvandy Haris,Rizky Rahmansyah. 2026. “The Application of Naive Bayes in Analyzing Public Sentiment Toward the Performance of the North Sumatra Regional Government in Handling Flash Floods”. Formosa Journal of Computer and Information Science, Vol. 5 No. 1 Tahun 2026.
DOI: https://doi.org/10.55927/fjcis.v5i1.16417
URL: https://journal.formosapublisher.org/index.php/fjcis