A team of University of Waterloo researchers has developed a new machine learning method that detects hate speech on social media platforms with 88 percent accuracy, saving employees hundreds of hours of emotionally damaging work. is saved from
The method, dubbed the Multimodal Discussion Transformer (mDT), can understand the relationship between text and images as well as comments in more context than previous hate speech detection methods. can put This is particularly helpful in reducing false positives, which are often falsely flagged as hate speech due to culturally sensitive language.
“We really hope this technology can help humans reduce the emotional costs of manually filtering out hate speech,” said Liam Hibbert, a Waterloo computer science PhD student and first author of the study. ” “We believe that by taking a community-based approach to our applications of AI, we can help create safer online spaces for everyone.”
Researchers have been building models to analyze the meaning of human discourse for many years, but these models have historically struggled to understand meaningful conversations or contextual statements. Previous models have only been able to identify hate speech with 74 percent accuracy, lower than Waterloo's research.
Hebert said context is crucial to understanding hate speech. “For example, the comment 'That's gross!' It may be innocuous by itself, but its meaning changes dramatically if it is in response to a picture of a pizza with pineapple versus a member of a disadvantaged group.
“Understanding this distinction is easy for humans, but training a model to understand the context in a discussion, including considering images and other multimedia elements within them, is actually a very difficult problem.”
Unlike previous efforts, the Waterloo team built and trained their model on a data set that contained not only isolated hateful comments but also the context of those comments. The model was trained on 8,266 Reddit discussions containing 18,359 labeled comments from 850 communities.
“More than three billion people use social media every day,” Hebert said. “The influence of these social media platforms has reached unprecedented levels. There is a great need to address hate speech on a large scale so that spaces can be created where everyone is respected and safe.”
The research, Multimodal Discussion Transformer: Integrating Text, Image and Graph Transformers to Detect Hate Speech on Social Media, was recently published in the Proceedings of the 38th AAAI Conference on Artificial Intelligence.