Public feedback analysis on multi-stage emergency management policies using BERTopic-SKEP integrated model

Area of Science:

• Disaster Management and Governance
• Computational Social Science
• Artificial Intelligence in Public Policy

Background:

• Effective disaster governance relies on understanding public feedback throughout multi-stage emergency management policies.
• Previous methods struggle with analyzing public sentiment and concerns in short-form social media data.

Purpose of the Study:

• To develop and apply an integrated deep learning framework for quantifying public perceptions of government actions during the "23.7" Beijing Rainstorm.
• To analyze public concerns and emotional responses across the four emergency management stages: prevention, preparedness, response, and recovery.

Main Methods:

• Utilized BERTopic for topic modeling and SKEP for sentiment analysis on 50,015 social media posts.
• Integrated semantic embeddings and structured sentiment knowledge to overcome limitations of traditional topic modeling.
• Applied the Narrative Policy Framework (NPF) to interpret findings as structured policy narratives.

Main Results:

• Public attention was heavily concentrated in the response and recovery stages (over 76%), with minimal focus on prevention and preparedness (under 2.5%).
• The recovery stage exhibited high positivity (78.18%), indicating public approval of post-disaster actions.
• NPF analysis identified public perceptions of the government as a "hero" or "planner" within policy narratives.

Conclusions:

• The integrated BERTopic-SKEP framework effectively captures nuanced public demands and emotional dynamics in emergency management.
• Findings provide actionable insights for optimizing policy legitimacy and operational efficacy in disaster governance.
• Understanding multi-stage public feedback is critical for enhancing disaster response and recovery efforts.