The problem of bullying has always been a social problem. Until the modern age, it tended to be a face-to-face issue that people faced, but in the age of always-on communications devices, cyber-bullying has emerged as a serious matter that must be addressed. New research in the International Journal of Computational Science and Engineering, looks at an ensemble approach to detecting incidences of cyber-bullying.
Pradeep Kumar Roy if the Indian Institute of Information Technology in Surat, Gujarat, Ashish Singh of the KIIT Deemed to be University in Bhubaneswar, and Asis Kumar Tripathy and Tapan Kumar Das if Vellore Institute of Technology in Vellore, India, point out that modern communications technology has so many advantages for society but as with all inventions comes with a flipside. The negative aspects of the many tools with use in our digital lives include crime, spam, and cyberbullying.
Given that more than half the world’s population is now active on the internet in some form via computers and mobile devices and a huge proportion of those use the many disparate social media sites as well as more conventional tools, such as email and the web, there is plenty of scope for the cyberbully to attack.
The team’s ensemble machine learning model examines online Twitter posts and uses a two-stage process to analyse content. The first step involves applying k-nearest neighbour, logistic regression and, decision tree classifiers. This is the underlying classification as to whether a post is bullying or not. But, the second stage involves precluding false positives and false negatives by applying a voting-based ensemble learning model to the classification. The team’s experiments with known data confirmed that the ensemble model can detect bullying posts with a good degree of accuracy, around 94 per cent.
Such a level of accuracy is sufficiently high that those who have oversight of accounts on various systems might be able to focus an examination of activity from a purported cyberbully so that follow-up decisions might be made in terms of limiting their accounts to preclude further bullying. Future work will involve extending the tools to other social media platforms. The potential for linking together data from more than one such system might allow even greater accuracy to be achieved so that cyberbullies operating across platforms might be identified.
Roy, P.K., Singh, A., Tripathy, A.K. and Das, T.K. (2022) ‘Cyberbullying detection: an ensemble learning approach’, Int. J. Computational Science and Engineering, Vol. 25, No. 3, pp.315–324.