This research develops an AI system using att-Bi-LSTM to detect community safety crises in news and social media. It processes both text and voice in Amharic and English with high accuracy and F1 scores. The model aids quick response by directing critical reports to relevant organizations. Its attention mechanism improves detection performance, even in low-resource languages. The work lays the groundwork for scalable, multilingual crisis monitoring solutions.

Fig. 8. Comparison of accuracy between att-Bi-LSTM and classical models.

Technology Overview
This study proposes a deep learning framework called attention-based Bidirectional Long Short-Term Memory (att-Bi-LSTM) to detect Community Safety Crisis Incidents (CSCI) in digital news reports. It leverages both text and speech data from platforms such as Twitter, YouTube, and news websites. The model is evaluated on Amharic and English datasets, demonstrating high accuracy and F1 scores across languages. The architecture integrates attention mechanisms and fastText word embeddings, outperforming classical machine learning models and attention-less networks. The framework supports both text-based pre-training and speech recognition-based detection, making it adaptable to multimodal input sources.

Applications & Benefits
The proposed system enables real-time identification of crisis-related content, helping organizations respond faster to community safety threats. With accuracies up to 90.93% for text and 82.10% for voice, it supports automated monitoring of reports in low-resource languages like Amharic. The model's multilingual adaptability allows cross-language application, expanding crisis detection in underrepresented regions. Its use of attention-based deep learning improves reliability in identifying and classifying critical incidents. Future enhancements, including multi-class classification, will help differentiate between various crisis types, increasing actionability for emergency and humanitarian responders.

How can we detect news surrounding community safety crisis incidents in the internet? Experiments using attention-based Bi-LSTM models
Author：Yeshanew Ale Wubet, Kuang-Yow Lian
Year：2024
Source publication：International Journal of Information Management Data Insights, Volume 4, Issue 1, April 2024
Subfield Highest percentage：99%  Library and Information Sciences  #2 / 287

https://www.sciencedirect.com/science/article/pii/S2667096824000168