Kanlun (Sampson) Wang, PhD, assistant professor of analytics at Fairfield University’s Charles F. Dolan School of Business, has published a research paper titled “Digital Voices of Survival: From Social Media Disclosures to Support Provisions for Domestic Violence Victims.” The study explores how artificial intelligence can help researchers better understand how individuals disclose experiences of domestic violence on social media—and how online communities respond.
Dr. Wang explained that disclosures can include any post or comment on a social media platform in which an individual publicly shares that they have experienced some form of domestic violence. Existing research in this area focuses primarily on detecting posts related to abuse; Dr. Wang’s work takes the analysis one step further.
“The research extends beyond merely detecting instances of domestic violence in online communities,” he said. “Rather than simply identifying abuse online, the study investigates community responses and explores how platforms can more effectively support victims.”
His research maps the relationship between the types of abuse described and the types of support offered by online communities, providing insight into how people respond when individuals seek help online. Ultimately, his goal is to move toward stronger intervention and support mechanisms for those seeking help through digital platforms.
The Impact of Anonymity
Emphasizing a careful and responsible approach to analyzing such sensitive content, Dr. Wang analyzed thousands of posts from platforms including Reddit and Facebook. He found that Reddit’s largely anonymous structure often leads to longer, more detailed posts, while Facebook conversations may involve different forms of community interaction.
AI-Driven Detection & Support Provisions
Dr. Wang recently presented his latest work at the Hawaii International Conference on System Sciences. Using an AI-driven framework to identify self-disclosed instances of domestic violence, he reported that his team achieved nearly 80 percent accuracy. The model also maps community-based interventions for victims in need of timely support, capturing both positive and negative signals in online discourse.
