Sentiment Analysis for Disaster Risk Management Using Data from the Pakistan Citizen Portal: A Case Study on Flood and Heatwave Complaints

Abstract

Effective disaster risk management requires bottom-up approaches that leverage citizen feedback to gain insights into disaster events on field. While much existing research in this domain focuses on social media platforms, the rich potential of government-operated digital feedback systems in capturing citizen experiences and concerns during disaster events remains underexplored. To address this gap, this study utilizes sentiment analysis to evaluate complaints submitted on the Pakistan Citizen Portal concerning flood and heat events. A hybrid methodology is adopted, combining VADER, SVM, LR, and BERT models to uncover sentiment trends and emotions expressed by citizens. Additionally, Latent Dirichlet Allocation is employed to identify key topics of concern, enhancing decision making and resource allocation. Our findings demonstrate the effectiveness of BERT in capturing sentiment trends while addressing the linguistic and cultural nuances inherent to the data. Emotions are detected based on Ekman’s six basic emotions, with sadness and anger emerging as the predominant ones. Key topics identified include property damage from floods and electricity problems related to heat events. The insights derived from this analysis provide a policy-relevant assessment of disaster risk management in Pakistan, illustrating how digital citizen feedback can inform resource allocation, preparedness planning, and communication strategies. By translating citizen sentiment into actionable insights, this study demonstrates the potential of digital platforms like the Pakistan Citizen Portal to enhance community resilience, strengthen preparedness, and support more responsive and inclusive disaster governance.