CSAD-CYBER SECURITY HYBRID LEARNING MODELS FOR DETECTING AGGRESSION IN SOCIAL MEDIA

Abstract

Cyber security data aggression detection (CSAD) is the most important in social media. Unquestionably, people rely too heavily on social media to communicate effectively. However, there is no suitable restriction on who can participate in communication. Therefore, anonymous senders of irrelevant and occasionally hostile messages undermine the core goal of successful communication. Its influence on society grows along with its popularity, shifting from largely positive to negative. Online aggression, which referred to as the deliberate information use technology to harass, threat, defamation, or otherwise damage a further party, has a negative effect. There is a growing need for automatic filters to find and delete these undesired messages due to the volume of tweets, text messages, and other forms of rewets. However, the majority of current approaches only take into account NLP-based feature extractors, such as TF-IDF and Word2Vec, without taking into account emotional aspects, making them less efficient for detecting cyber violence. In this study, we retrieved eight new emotional variables to recognize hostile remarks using the newly created a three-layer deep neural network. Dataset for Cyber-Troll and ASR was used to test the suggested HDNN design. Embedding words and 8 different emotional variables mixed together and fed into to the HDNN significantly increase recognition while maintaining a straightforward and computationally less taxing approach. Our suggested model outperforms the rivals by a wide amount, scoring an F1 score of 97% when compared to the most recent models.