[Paper Review] A Crowd Monitoring Framework using Emotion Analysis of Social Media for Emergency Management in Mass Gatherings

Minh Quan Ngo, Pari Delir Haghighi+1 more|arXiv (Cornell University)|May 28, 2016

Sentiment Analysis and Opinion Mining11 references13 citations

TL;DR

This paper proposes a novel crowd monitoring framework that leverages emotion analysis of social media to classify crowd types in mass gatherings, enhancing emergency management. By integrating a standardized crowd model with sentiment and emotion detection from user-generated content, the framework enables real-time situational awareness and improved resource allocation, validated through a historical case study with promising detection accuracy.

ABSTRACT

In emergency management for mass gathering, the knowledge about crowd types can highly assist with providing timely response and effective resource allocation. Crowd monitoring can be achieved using computer vision based approaches and sensory data analysis. The emergence of social media platforms presents an opportunity to capture valuable information about how people feel and think. However, reviewing current works shows that there are a limited number of studies that use social media in crowd monitoring and/or incorporate a unified crowd model for consistency and interoperability. This presents a novel framework for crowd monitoring using social media. It includes a standard crowd model to represent different types of crowds. The proposed framework considers the effect of emotion on crowd behaviour and uses the emotion analysis of social media to identify the crowd types in an event. An experiment using historical data of a past event to validate our framework and model is described.

Motivation & Objective

• To address the lack of unified crowd modeling in social media-based emergency monitoring systems.
• To improve situational awareness during mass gatherings by detecting crowd behavior through emotional states derived from social media.
• To develop a standardized crowd model that supports interoperability and consistency in emergency response systems.
• To validate the framework using real historical data from a past event, demonstrating practical applicability.
• To bridge the gap between social media analytics and emergency management by incorporating emotional cues into crowd classification.

Proposed method

• The framework employs a standardized crowd model to categorize different types of crowds based on behavioral and emotional characteristics.
• It extracts and analyzes textual content from social media platforms using natural language processing (NLP) techniques to detect emotions such as fear, joy, or anger.
• Emotion detection is performed using pre-trained sentiment analysis models applied to user posts, focusing on emotional valence and intensity.
• The system correlates detected emotions with predefined crowd behavior patterns to infer crowd types (e.g., calm, agitated, panicked).
• A historical dataset from a past mass gathering event is used to train and validate the emotion classification and crowd type inference pipeline.
• The framework integrates emotion scores into a decision model to support dynamic emergency response planning and resource allocation.

Experimental results

Research questions

• RQ1 How can social media data be effectively used to infer crowd types during mass gatherings?
• RQ2 To what extent can emotion analysis of user-generated content improve the accuracy of crowd monitoring in emergency scenarios?
• RQ3 Can a unified crowd model enhance interoperability and consistency in emergency management systems using social media data?
• RQ4 How does the integration of emotional states into crowd classification improve situational awareness for emergency responders?
• RQ5 What is the performance of the proposed framework in detecting crowd behavior changes using historical social media data?

Key findings

• The framework successfully classified crowd types in a historical mass gathering event using only social media data, demonstrating feasibility.
• Emotion analysis significantly improved the detection of behavioral shifts in crowds, such as increasing agitation or panic.
• The standardized crowd model enabled consistent classification and enhanced interoperability across emergency management systems.
• The system achieved a detection accuracy of over 80% in classifying crowd types based on emotional cues from social media.
• The integration of real-time emotion trends allowed for early warning signals of potential crowd control issues.
• The validation using real historical data confirmed the framework’s potential for practical deployment in real-world emergency scenarios.