Visual Sentiment Analysis from Disaster Images in Social Media

Area of Science:

• Computer Science
• Artificial Intelligence
• Social Media Analysis

Background:

• Social media fuels sentiment analysis with diverse content, yet visual sentiment analysis, especially for disaster events, remains underexplored.
• Analyzing public sentiment in disaster imagery is crucial for humanitarian aid and disaster response.
• Existing research primarily focuses on text-based sentiment analysis, leaving a gap in visual data interpretation.

Purpose of the Study:

• To develop and evaluate a deep visual sentiment analyzer specifically for disaster-related images.
• To create a large-scale, publicly available benchmark dataset for visual sentiment analysis in disaster contexts.
• To provide a foundational resource for future research in social media disaster analysis.

Main Methods:

• A deep learning model was designed for visual sentiment analysis of disaster imagery.
• A global crowd-sourcing study was conducted for data annotation and sentiment labeling.
• A benchmark dataset with multiple annotation sets was created for various analytical tasks.

Main Results:

• A novel deep visual sentiment analyzer for disaster images was successfully developed.
• A comprehensive benchmark dataset was generated through a large-scale crowd-sourcing effort.
• The study established a baseline for visual sentiment analysis in the domain of social media disaster response.

Conclusions:

• The proposed deep visual sentiment analyzer and benchmark dataset advance the field of visual sentiment analysis in disaster contexts.
• This research offers valuable tools for stakeholders like humanitarian organizations and news broadcasters.
• The findings contribute to building more resilient communities by improving disaster information analysis.