An Empirical Methodology for Detecting and Prioritizing Needs during Crisis Events

Area of Science:

• Natural Language Processing
• Crisis Informatics
• Social Media Analysis

Background:

• Identifying public needs during crises is vital for resource allocation.
• Social media platforms like Twitter offer valuable, yet noisy, data for needs assessment.
• Challenges include information sparsity and irrelevant content in crisis-related social media data.

Purpose of the Study:

• To develop novel methods for detecting needs from social media during crises.
• To extract lists of required resources (e.g., masks, ventilators).
• To identify sentences specifying resource requirements and beneficiaries (who-needs-what).

Main Methods:

• Proposed two distinct natural language processing methods for needs detection.
• Applied methods to a dataset of COVID-19 crisis-related tweets.
• Evaluated resource extraction against official lists and sentence detection against annotated tweets.

Main Results:

• Achieved 0.64 precision in extracting lists of needed resources.
• Obtained a 0.68 F1-score for detecting who-needs-what sentences.
• Demonstrated effectiveness in identifying critical information from noisy social media data.

Conclusions:

• The developed methods offer a viable approach for real-time needs assessment during crises.
• Automated social media analysis can significantly improve crisis resource management.
• Future work can refine these methods for broader crisis communication applications.