MSSThreD: Multitask Social Media Sentiment Analysis with Sarcasm for Public Security Threat Detection

Public security threat detection is critical for maintaining safety of citizens from possible natural or non-natural disasters. Social media, with its vast and diverse content, serves as a significant source for communication and information dissemination. Early detection of threats through social media analysis can greatly benefit in prevention before the event, or response and rescue during the event. However, the noisy textual with linguistic nature of social media content presents substantial challenges for sentiment analysis. Coupling with sarcasm can reverse the polarity of sentiments, complicating threat detection efforts. To address these challenges, we introduce an application framework for public security threat detection that embeds multitask learning, incorporating sentiment analysis and sarcasm detection. This application utilized preprocessing and feature engineering techniques based on Natural Language Processing to capture the nuance of sentiment and sarcasm, with deep learning network model of Bidirectional Long Short-Term Memory and Gated Recurrent Unit combination. The sentiment filtering and identifying threat-related keywords, producing monitoring reports that highlight threats earlier than traditional web news sources, as evidenced by our case study results. This application demonstrates the practical utility and impact of the framework in enhancing public security through timely threat detection.