Research Information System
Turkiye Klinikleri Journal of Health Sciences, vol.4, pp.1-5, 2024 (Peer-Reviewed Journal)
Objective: The frequency of natural disasters worldwide
is increasing, and social networks have become popular sources of crucial data for analyzing images’ emotions. Although the analysis of disaster-related images is a relatively new field, this study aims to identify
the emotional responses evoked by images shared on social media. Material and Methods: In this four-stage study, a total of 5,203 free and
openly accessible images were scraped from various social media platforms, and emotion categories associated with these images were selected. The images were converted to RGB format and resized after
undergoing preprocessing. Normalization of the visual pixels was performed. Various deep learning (DL) models were examined for visual
sentiment analysis, and their performance was compared using metrics.
Subsequently, emotion classification was performed using Inception-v3,
which yielded the most reliable results. Results: The most suitable DL
model among different pre-trained DL models was determined to be Inception-v3 with a performance metric of 81.2%. The analysis of the emotions depicted in the images revealed that 71.9% (n=3,741) were
classified as negative, while 8.0% (n=781) were classified as neutral.
Conclusion: These results indicate that visual sentiment analysis of social media data can significantly enhance disaster response efforts. By
identifying early warning messages, updating disaster-related information, and monitoring user-generated content, this approach supports more
effective data analytics and information dissemination. Consequently,
the use of advanced DL models like Inception-v3 in analyzing emotional
content from social media can provide valuable insights and improve the
efficiency and effectiveness of disaster management strategies.