Hate speech and misinformation on social media can have devastating effects, especially in marginalized communities. But what if we used artificial intelligence to combat such harmful content?
That’s the goal of a team of University of Toronto researchers who won a catalytic grant from the Data Science Institute (DSI) to develop AI systems that address community marginalization in data-centric systems, including social media platforms like Twitter. .
The joint research team consists of the following members. Syed Ishtiak AhmedAssociate Professor in the Department of Computer Science, Faculty of Humanities and Sciences. Shohini Bhattasari, T Assistant Professor of Linguistics, University of Scarborough.and Shion GuhaI am an assistant professor in both the computer science and informatics departments, and the director of the human-centered data science laboratory.
Their goal is to make content moderation more inclusive by involving communities affected by harmful or hateful content on social media. The project is a collaboration with two of his non-profit organizations in Canada, the Chinese Canadian National Social Justice Council (CCNC-SJ) and the Islamic Unraveling Anti-Racism Initiative.
Historically marginalized groups are most vulnerable to content moderation failures because they are less representative among human moderators and less data is available for algorithms, Ahmed explained. .
(LR) Syed Ishtiak Ahmed, Shohini Bhattassari, Shion Guha (Photo provided)
“While most social media platforms manage and identify harmful content and take steps to limit its spread, human moderators and AI algorithms can correctly identify it and take appropriate action. It often fails,” he says.
The team will design and evaluate the proposed system to address potential Islamophobic and Chinaphobic posts on Twitter. The AI system primarily aims to democratize content moderation by incorporating diverse opinions in two ways. First, by allowing users to challenge decisions, the moderation process becomes more transparent and trustworthy for users who are victims of online harm. Second, the system takes user input and retrains the machine learning (ML) model to ensure that the user’s competitive position is reflected in the pre-screening ML system.
“Different annotators make it difficult to annotate data. Solving this problem democratically requires involving different communities, which is not currently common in data science practice.” No,” points out Ahmed.
“This project is working with two historically marginalized communities in Toronto to address this issue by designing, developing and evaluating a multidimensional framework of justification and contention in data science. .”
AI systems integrate the knowledge and experience of community members into the process of curtailing hateful content directed at the community. The team uses participatory data curation techniques that help them learn the characteristics of the different types of harmful content that affect their communities, and they go through the data labeling process to ensure data quality. Engaging members of the corresponding community.
“We are grateful to DSI for their generous support of this project. Thanks to the DSI community, we have been able to connect with and learn from people doing similar research,” said Ahmed, adding that his team The research is expected to have far-reaching impacts beyond the two communities we are currently focusing on, he added.
