Extracting Actionable Requirements from Crisis Event Tweets for Requirements Engineers

Abstract

Developing crisis-related software applications is a key part of crisis management, but gathering comprehensive requirements for the application is a challenge for requirements engineers, as incomplete requirements can lead to applications lacking essential features. With the rise of social media, vast amounts of public data are now available for extracting crisis-related data to be used for feature specification. However, extracting key information poses challenges. This study therefore introduces a framework for extracting actionable requirements from social media, specifically analyzing crisis related tweets using fine-tuned machine learning models. The framework identifies highly relevant tweets based on a taxonomy of sources of information and the priority of tweets, then employs tools and techniques like word clouds, emotion analysis, and topic modeling to reveal insights. We used a 2-2-step prompting approach with GPT-4o to extract requirements through actionable tweets. In our study, 87 requirements were extracted from the actionable information sources, reflecting the needs expressed during crises. These requirements were then translated into potential features for a "Crisis or Emergency Management" application. When compared with 32 existing mobile apps, it shows a 75.6% match and identifies 34 unique features for future development.