Integrating AI and Co-production to Analyze Communications in Social Media Substance Use Recovery Groups

About This Grant

Substance use disorder (SUD) is a leading public health challenge, with long-term consequences for physical and mental health. In-person peer support groups are well-established as beneficial for recovery. However, as digital platforms increasingly serve as spaces for peer support, little is known about how engagement in online peer support recovery groups is associated with recovery outcomes. Emerging research has yielded conflicting results, with some studies suggesting benefits while others indicating relapse risks. This underscores the need to examine the content of discussions beyond engagement frequency. The present application seeks support for Xiangyu Tao, Ph.D., a postdoctoral associate at the Rutgers Addiction Research Center, gain the necessary training and set up a line of research to examine the role of online peer support recovery communities in SUD recovery. Specifically, she will integrate state-of-the-art artificial intelligence (AI) techniques, including Large Language Models (LLMs), with co-production to identify communication patterns and to examine their associations with recovery outcomes. LLMs offer a promising avenue for analyzing large-scale online discussions, yet they require human oversight to address challenges such as contextual misinterpretation and ethical concerns. Co-production, i.e., involving individuals with SUD recovery experience in all research stages, mitigates LLM limitations and ensures that results reflect lived experience. The mentored K99 phase will identify and characterize communication patterns in online recovery groups. Peer support and co-rumination patterns will be classified using LLMs and co-production (Aim 1); latent class analysis (LCA) will identify distinct communication profiles among users engaging in these online groups (Aim 2). During this phase, Dr. Tao will receive mentorship in co-production and AI methodologies, longitudinal data analysis and management, and responsible AI research. The independent R00 phase will build upon this foundation by examining how communication profiles are associated with recovery trajectories using longitudinal survey data from online recovery group participants (Aim 3). This project is highly innovative in its integration of LLMs and co- production to analyze large-scale digital recovery discussions, ensuring that AI-driven insights are both computationally rigorous and socially informed. Findings will enhance understanding of digital peer support for SUD recovery and will inform future mechanistic studies and SUD interventions. By identifying communication patterns associated with recovery trajectories, this project will guide digital health platforms, peer support programs, and clinicians in optimizing online recovery environments to better support individuals with SUD. This project aligns with the NIDA Strategic Plan Cross-Cutting Priority to "Leverage Data Science and Analytics to Understand Real-World Complexity" by utilizing advanced computational methods to analyze digital recovery support interactions. The project is highly significant in bridging advanced AI computational tools with the lived experiences of individuals to produce impactful research to promote substance use recovery.