Deep Learning Framework for Prior Identification of Threats in Social Media Interactions

Abstract: Social Media Interactions are digital platforms that enable users to create profiles, connect with others, and interact through various forms of communication. These platforms facilitate social interaction, content sharing, and collaboration across geographical boundaries. Threats in Social media interactions refer to risks and malicious activities that exploit vulnerabilities in online platforms, posing harm to users, data, and digital ecosystems. These threats can impact personal security, privacy, and overall platform trustworthiness. The Deep Learning-Based Framework for Prior Identification of Threats (DLPIT) is designed to proactively detect and mitigate harmful content in social media interactions, such as hate speech, cyberbullying, and violent language, before they escalate. By leveraging a Recurrent Neural Network (RNN), which is particularly effective for sequential data processing, the system analyzes Twitter language and classifies information as either harmful or non-threatening. The framework is trained on a preprocessed labeled Twitter dataset that incorporates both textual and behavioral data, ensuring comprehensive threat detection. The RNN's ability to capture contextual relationships and temporal dependencies enables DLPIT to monitor social media platforms in real time with high efficiency. Furthermore, the framework enhances detection accuracy by integrating social network interactions and user engagement patterns, which help in identifying the potential influence and reach of harmful content. To quantify the severity of a detected threat, the system calculates a Threat Level Score (TLS) based on multiple factors, including the intensity and frequency of harmful words, user history, past engagement patterns, and the influence of the content within the social network. A higher TLS signifies a greater risk, enabling moderators to prioritize intervention and take necessary actions accordingly. The performance of DLPIT is rigorously evaluated and compared with existing methods using F1-score, recall, accuracy, and precision.

Keywords: Recurrent Neural Network (FFNN), Threat Level Score, Explainable AI, Cyber Bulling