Researchers Develop New Method for Detecting Hate Speech
Associate Professor Marian-Andrei Rizoiu of the University of Technology Sydney is leading research to combat online hate speech, which can deepen political divides, marginalize vulnerable groups, and pose risks to democracy and public safety. His interdisciplinary approach combines computer science and social science to understand how online speech influences public opinion.
Rizoiu’s team developed a multi-task learning (MTL) model, which can handle multiple tasks simultaneously and apply insights across datasets. This model, described in their study on hate speech detection involving political figures, was trained on eight hate speech datasets from platforms like Twitter, Reddit, Gab, and Stormfront. It was then tested on 300,000 tweets from American public figures across the political spectrum.
The findings showed that most hate-filled and abusive content, often related to misogyny and Islamophobia, came from right-leaning individuals, with 5093 out of 5299 abusive posts linked to such figures. The model effectively differentiated between hate speech and other abusive content, identifying topics like religion, gender, ethnicity, and immigration. This approach highlights the need for nuanced, adaptable models to accurately detect and address diverse forms of online hate speech.