Description

Social media platforms (SMPs) are essentially transforming the way we communicate, collaborate, consume, and create information. These have had an incredible positive effect on freedom of expression and information, facilitating public discussion, sharing of information and strengthening civic space. But it puts publishing in the hands of the masses, without sufficient armour and expression regulations. Moreover, it makes tailing, profiling, and targeting overly accessible. Further, the absence of temporal limits, imperishability and virality of online content creates nuisance. Undeniably, the low detection and conviction rate makes social media accounts vulnerable to messages of an intimidating or threatening nature. Certainly, online abuse in SMPs has become a pervasive problem reflective of underlying societal inequalities and divisions. But abuse is a generic term which encompasses most of the virtual hate and abuse crimes which have real-time consequences. There clearly existed a gap in congregating the diversity of techniques used for automatic detection of abuse, hate or bullying in the user-generated big data content on social media. Therefore, the primary objective of this research is to understand the diverse concept of abuse in terms of which online activities on social media qualify as abusive behaviour and which in-person abusive activities can be captured using cyber-physical systems. It is identified that online abusive behaviour can be in many forms and denigration is a specialized form of cyberbullying which describes a recurrent, sustained and intentional attempt to damage the victim's reputation or ruin the friendships that he or she has by spreading unfounded gossip or rumours online. It is the most common bullying tactic involving character assassination of public figures like celebrities and politicians. Conventional mechanisms to counter denigration cyberbullying consist of using standard guidelines, human moderators, and vetoes based on the use of abusive/ offensive words. However, a collective approach is required to match to the scale of social media which can automatically detect denigration cyberbullying. Recently, soft computing techniques have been widely used as a valuable and feasible solution to the emerging social media and big data. This research presents a variety of mechanisms to counter denigration in online posts by debunking reputation rumours. Various hybrid models (Lexicon-based + machine learning; Lexicon + filter-wrapper machine learning; deep learning + filter-wrapper machine learning) and swarm-optimized machine learning model are proposed for denigration bullying detection in social media content. Lastly, a text-based analytical model for automatic denigrate comment detection in low-resource language, Hindi is put forward. Thus, the key contribution of the research is a comprehensive approach to tap the real-time dynamics of social media for automatic detection of denigration cyberbullying in online textual content using soft computing techniques. The results are evaluated and compared with various baselines using synthetic and benchmark datasets. The preliminary results are motivating as the models comprehend the complexities of natural language in online social media content where the representations of textual data is learned as real-valued vectors. As promising future direction, models and benchmark datasets for multimodal denigration bullying detection are desired.