Improving Disaster Response Through Hierarchical Classification of Emergency Text Data Using Mamba State Space Model

View at publisher Summarise with VIRAT
Authors: Harsha Prabha V S, Agveena Xavier, Janani M
Cited by: 0
References: 14

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Disaster scenarios generate huge amounts of short and unstructured text messages over SMS, social media, and emergency reporting systems with huge variations. These will often have critical information in them about injuries, people missing, infrastructure failures etc as well as requesting urgent assistance. However, the review and organization of such data through manual means is extremely time consuming, open to human error, and quite impossible in life- threatening situations where every minute counts. This paper proposes a hierarchical text classification framework completely made on a very recent text classification architecture of highly efficient deep learning for fast sequence processing called Mamba SSM. Unlike the classical Transformer-based models with great computational costs and slow-time inferences (quadrature time complexity), Mamba has linear complexity and a selective state mechanism that can remember the important information while filtering the noise, making it very suitable for real-time disaster data processing. The proposed system is trained on the Figure Eight disaster response dataset, which contains more than 26,000 real-world messages annotated in 36 categories corresponding to major types of disasters and certain aid needs. By extensive preprocessing and fine- tuning, the model learns to automatically classify messages into multiple relevant labels at the same time, this enables capturing the hierarchical structure of emergency information. Experimental evaluation, the Mamba-based model achieves good precision, recall, F1 scores in most categories, and can also be significantly better than traditional architectures in terms of inference speed. The model can determine 1000 messages in a few seconds, showing that it can be practically used in the real time of the emergency. The framework converts noisy and disorganized text to structured and actionable insights to help with situational awareness of emergency responders more accurately. It helps authorities in the prioritization of critical cases, managing resources more effectively, and quick response towards needy people. Overall, this research underlines the potential that the MAMBA architecture has for modern disaster management in terms of power and scalability. Future enhancements might involve handling rare category imbalance, integrate multilingual data, or combine geospatial and temporal features for even stronger decision-making support.

Fields of Study & Subjects

Text and Document Classification Technologies Topic Modeling Public Relations and Crisis Communication

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
1562–1567
Publication date
21 January 2026
Language
EN