Nayar Prize II , Phase I Quarterly Progress Report

Highlights: The key accomplishments achieved during Phase I of this Nayar Prize II project are as follows: (1.) Conducted dozens of interviews with parents, school administrators, and law enforcement to understand where cyberbullying takes place, what types of cyberbullying are the most important to detect, who is the most appropriate person to alert, and how the output of a forecasting system should be communicated to relevant stakeholders. (2.) Collected more than 15 million comments from more than 400,000 Instagram posts, of which more than 30,000 comments have been manually annotated for the presence of cyberbullying content to train and validate the forecasting model. (3.) Designed and implemented several competing machine learning models to forecast the presence and intensity of cyberbullying in Instagram posts. The best method is able to predict the appearance of a hostile comment on a post ten or more hours in the future with 77% accuracy and can furthermore distinguish between high and low levels of future hostility with 87% accuracy. (4.) Designed and implemented a web-based user interface that allows parents to monitor the Instagram accounts of their children’s online community and receive alerts of imminent cyberbullying threats. The prototype is accessible at http://nayar.casmlab.org with the username “[email protected]” and password “password”. (5.) Submitted one NSF proposal and one computer science research article based on this work, with a second NSF proposal under development.