Using AI to detect hate speech in context

Thursday, May 30, 2024

A team of researchers at the University of Waterloo have developed a new machine-learning method that detects hate speech on social media platforms with 88 per cent accuracy, saving employees from hundreds of hours of emotionally damaging work.

The method, dubbed the Multi-Modal Discussion Transformer (mDT), can understand the relationship between text and images as well as put comments in greater context, unlike previous hate speech detection methods. This is particularly helpful in reducing false positives, which are often incorrectly flagged as hate speech due to culturally sensitive language.

“We really hope this technology can help reduce the emotional cost of having humans sift through hate speech manually,” said Liam Hebert, a Waterloo computer science PhD student and the first author of the study. “We believe that by taking a community-centred approach in our applications of AI, we can help create safer online spaces for all.”

Read the full article from Waterloo News to learn more.