Topic Modeling and Sentiment Classification: Perspectives From Libyan Flood Conversations on Social Media
Year: 2026
Type: article
Abstract: Social media streams related to disasters are a key but difficult source of data to use in real-time situational awareness. This study presents a merged sentiment analysis model, which is based on classical machine learning, transformer models, and a metaheuristic optimization approach, which integrates theSalp Swarm Optimization(SSO) algorithm. The suggested ensemble uses SSO to optimize the hyperparameters and decision-fusion weights of a harmonized model with a balance between interpretability, computational resources, and predictive strength. Extensive experiments along historical baselines (1982–2024) have shown that the SSO-improved ensemble achieves state-of-the-art performance with 97% accuracy, which is superior to both of the individual deep models and previously used ensemble techniques. Quantitative findings indicate better speed-accuracy trade-offs and cost efficiency, whereas qualitative findings indicate that the model can reflect sentiment change in four phases of disaster. Experiments on cross-disaster generalization also support the high level of transferability with an average accuracy of 90.7% in various crisis events. The results demonstrate that optimization-based ensemble design has provided a methodical and ethically viable route for real-time humanitarian monitoring, changing the social media sentiment into action-related crisis intelligence.
Source: IEEE Access
Authors: Ebtesam Mus A. Baka, Oğuz Ata
Institution: Altınbaş University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,075
OpenAlex ID: W7140282898