How can we detect news surrounding community safety crisis incidents in the internet? Experiments using attention-based Bi-LSTM models

Abstract

Reports related to community safety crisis incidents are being escalated and shared on social media and other online digital platforms. These reports must be addressed quickly to concerned organizations to provide welfare support to individuals and communities in crisis, to protect their lives, and to obtain justice. To achieve this, we proposed a framework termed detection of Community Safety Crisis Incidents (CSCI) reports using an attention-based Bidirectional Long Short-term Memory (att-Bi-LSTM). Amharic reports in Ethiopia were selected as the object of study for the implementation of the detection model due to the high CSCI report rate in the region. We gathered the textual data and spoken speech content reports from famous worldwide media websites, Twitter, and YouTube platforms utilizing data crawling techniques. The proposed model achieved 90.93 % accuracy for the text test dataset and 82.10 % accuracy for the voice test dataset on the text-based pre-training model. The model was also tested on English news, yielding an accuracy of 85.92 %.