Aggression Detection in Social Media from Textual Data Using Deep Learning Models

Abstract

It is an undeniable fact that people excessively rely on social media for effective communication. However, there is no appropriate barrier as to who becomes a part of the communication. Therefore, unknown people ruin the fundamental purpose of effective communication with irrelevant—and sometimes aggressive—messages. As its popularity increases, its impact on society also increases, from primarily being positive to negative. Cyber aggression is a negative impact; it is defined as the willful use of information technology to harm, threaten, slander, defame, or harass another person. With increasing volumes of cyber-aggressive messages, tweets, and retweets, there is a rising demand for automated filters to identify and remove these unwanted messages. However, most existing methods only consider NLP-based feature extractors, e.g., TF-IDF, Word2Vec, with a lack of consideration for emotional features, which makes these less effective for cyber aggression detection. In this work, we extracted eight novel emotional features and used a newly designed deep neural network with only three numbers of layers to identify aggressive statements. The proposed DNN model was tested on the Cyber-Troll dataset. The combination of word embedding and eight different emotional features were fed into the DNN for significant improvement in recognition while keeping the DNN design simple and computationally less demanding. When compared with the state-of-the-art models, our proposed model achieves an F1 score of 97%, surpassing the competitors by a significant margin.