Cyberbullying Detection Using Social Network Anlaysis
PRESENTER:
Qianjia Shy Huang
University of Ottawa
ABSTRACT:
This talk will give a brief description of a previous research for
cyberbullying detection and some ideas/trends for its future study.
Cyberbullying is an important social challenge that takes place over a
technical substrate. Thus it has attracted research interest across both
computational and social science research communities. While the social
science studies conducted via careful participant selection have shown the
effect of personality, social relationships, and psychological factors on
cyberbullying, they are often limited in scale due to manual survey or
ethnographic study components. Computational approaches, on the other
hand, have defined multiple automated approaches for detecting
cyberbullying at scale, but mostly have only focused on the textual
content of the messages exchanged. Unifying the two perspectives, the
researchers investigated a holistic (social +text) approach for
understanding and detecting cyberbullying. By analyzing the social
relationship graph between users in an online social network and deriving
features such as number of friends, network embeddedness, and relationship
centrality, we found that: (1) multiple social characteristics are
statistically different between the cyberbullying and non-bullying groups,
thus supporting many, but not all, of the results found in previous
survey-based bullying studies; and (2) analyzing such social network
features surrounding the network can yield significant improvements in the
accuracy of cyberbullying detection models as compared to purely
text-based models. Finally, the newer types of cyberbullying from
diversity social media and the current dataset from Instagram will be
discussed.