Phoenix, AZ – May 16, 2019 – With half of American teenagers reporting they have experienced some form of cyberbullying — from offensive name calling and spreading of false rumors to physical threats — a leading team of U.S. researchers is working to make Instagram, one of the most popular social media platforms among teens, a safer place to communicate. They’re doing this by combining math, computer science and psychology to develop a first-of-its-kind cyberbullying detection tool.
“With social networking, we get an amazing platform that enables us to communicate, exchange information and share opinions. But at the same time, it has become an environment that enables various kinds of damaging behavior,” said Yasin Silva, associate professor in the School of Mathematical and Natural Sciences at Arizona State University (ASU) and co-author of a paper presented at the recent Society for Industrial and Applied Mathematics (SIAM) International Conference on Data Mining entitled Hierarchical Attention Networks for Cyberbullying Detection on the Instagram Social Network.
According to a recent study by Pew Research Center, as many as 78 percent of American teens aged 15 to 17 used Instagram in 2018, making it an important platform to target. “Our ultimate goal is to develop online tools that can protect teenagers and better inform parents of undesirable activity,” Silva said.
The ASU cross-discipline research team represents a 50-50 distribution of researchers from psychology and computer science. A distinguishing characteristic of their work is the systematic application of research findings from psychology to inform their detection tool, ultimately providing more accurate identification of cyberbullying instances. Whereas other detection tools focus on analyzing text only, the ASU model successfully detects repetition of words, understands context, and mirrors the hierarchical posting structure of a complete Instagram session, taking into account the fact that comments occur at discrete times.
As Silva explained, the model uses a complex mathematical encoder to understand that original posts are followed by a sequence of comments at different time intervals, and that those comments may contain words that are damaging and require attention.
“Many psychologists agree that cyberbullying is something that doesn’t only happen once, but multiple times,” he said. “The goal of our model is to identify if an entire Instagram session is cyberbullying or not. It analyzes time intervals between comments to better capture the repetitive nature of cyberbullying, which further improves its detection accuracy.”
At the same time, the model leverages timing and social cues to understand words in context. For example, posting “I’m gay” is very different than a comment that uses the word gay in a slur, explained graduate research assistant Lu Cheng, a Ph.D. student in the ASU Department of Computer Science, and key member of the project team.
As another example, “If someone posts ‘What a great night!’ and someone else comments ‘Do the world a favor and go kill yourself,’ the last three words are relevant to cyberbullying and should receive more attention from a detection tool,” said Cheng. “But sometimes a word can have two different meanings based on context, and our encoder will capture that as well.”
Moving forward, the model will be used as the foundation for developing blocking apps and online resource tools. In September 2017, the ASU team announced a smartphone app called BullyBlocker that allows parents and victims of cyberbullying to monitor, predict and hopefully prevent incidents of online bullying on Facebook. Silva said a similar app is expected to be developed for Instagram, and will include a personalized set of links to relevant online resources in the event that a cyberbullying instance is detected.
“There are many ways that math and technology can play an active role to address the negative aspects of social networks,” he said. “The more our lives move into cyberspace, the more we need tools that can help us as both potential victims and parents to deal with these instances and receive help.”
About Society for Industrial and Applied Mathematics (www.siam.org)
Society for Industrial and Applied Mathematics (SIAM), headquartered in Philadelphia, Pennsylvania, is an international society of more than 14,000 individual, academic and corporate members from 85 countries. SIAM helps build cooperation between mathematics and the worlds of science and technology to solve real-world problems through publications, conferences, and communities like chapters, sections and activity groups. Learn more at siam.org.
To view the SIAM Journal article: https://bit.ly/30eL3ym