Machine Learning (ML) based Cyberbullying detection is a subject of rising interest amongst both computer science and psychology researchers. In this article, we present the results from a collaborative effort between the Faculty of Psychology (University of Lisbon) and Institute of System and Computer Engineering, Research and Development (INESC-ID), with the goal to develop a mobile application that can not only classify cyberbullying instances, but also teach and inform users to develop an assertive communication style that benefits their online experience. From data modelling and labelling, to building ML classifiers and implementing the “ComViver Online” app, we detail on our novel approach to have the data correctly represent the cyberbullying phenomenon, as well as study the effectiveness of the studied classifiers and the challenges that this task presents from a Natural Language Processing (NLP) point of view. By labelling the presence of cyberbullying through sets of interactions between Twitter users, as opposed to isolated tweets, we believe that is possible to capture the nature of the relation between mentioned users (friendly or not), as well as the notion of repetitiveness through a longer period of time, which is a defining feature of cyberbullying. Finally, we present the “ComViver Online” prototype app, which has been recently used by forty 9th grade students from schools in the Metropolitan Area of Lisbon, as a part of an intervention made in the scope of the following Portuguese Foundation for Science and Technology (FCT) funded project: PTDC/MHCPED/3297/2014.

--

Cyberbullying,machine learnin g,natural language processing,mobile application

• Computer and Information Sciences - Natural Sciences
• Electrical Engineering, Electronic Engineering, Information Engineering - Engineering and Technology
• Psychology - Social Sciences