Keynote #1
Title: Beyond Recommendations: Designing Conversational Music Agents for Self-Awareness and Psychological Well-being, by Li Chen
Abstract
As the demand for scalable mental health tools grows, the potential of Large Language Models (LLMs) lies in their ability to move beyond generic retrieval toward deeply adaptive, trustworthy support. This talk explores how music-driven conversational systems can bridge the gap between algorithmic recommendation and psychological intervention. The journey begins with establishing a foundation of trust through user-centric exploration. Using conversational music recommenders as a case study, I will discuss how critiquing strategies—such as progressive and cascading suggestions—shape user perceptions of helpfulness and serendipity. However, because trust is personal, I demonstrate how individual traits, including personality and trust propensity, dictate whether a user thrives under system-led or user-led initiatives.
Building on this rapport, the talk examines how these interfaces transition from suggesting songs to fostering psychological well-being. I will illustrate how music acts as a catalyst for self-awareness, using agents to guide users through emotional resonance and self-reflection. Furthermore, I will share findings on the role of generative AI in music-based reminiscence for older adults, where adaptive dialogue helps overcome cognitive barriers to memory recall. Collectively, these studies provide a roadmap for designing AI that moves beyond the playlist, transforming music-based chatbots into sustained partners for emotional and psychological health.
About Li Chen
Professor Chen specializes in human-centered AI, with a focus on developing reliable and trustworthy recommender systems from the users’ perspective. Her research moves beyond traditional accuracy metrics to prioritize serendipity, diversity, fairness, controllability, and explainability. By integrating insights from machine learning, psychology, and human-computer interaction, she explores user trust in recommender systems across diverse domains, including entertainment, e-commerce, social media, news, education, and mental well-being. She has authored and co-authored over 160 publications, most of which appear in high-impact journals (such as IJHCS, CSCW, TOCHI, TOIS, UMUAI, TIST, TIIS, KNOSYS, Behavior & Information Technology, AI Magazine, and IEEE Intelligent Systems), and key conferences in the areas of data mining (SIGKDD, WSDM, SDM), artificial intelligence (IJCAI, AAAI), information retrieval (SIGIR), recommender systems (RecSys), user modeling (UMAP), and intelligent user interfaces (SIGCHI, IUI, Interact).
Her co-authored papers have received several awards, including the RecSys'24 Best Student Paper Award, CHI'22 Honourable Mention Award, UMAP'20 Best Student Paper Award, UMUAI 2018 Best Paper Award, and UMAP'15 Best Student Paper Award. She received the President’s Award for Outstanding Performance in Teaching (Individual) 2024/25 and the President’s Award for Outstanding Performance in Research Supervision 2022/23. She has been included in the list of the world’s top 2% most-cited scientists by Stanford University since 2021, and the list of the Best Computer Science Scientists 2025/2026 by Research.com. She is now an ACM senior member, founding co-editor-in-chief of ACM Transactions on Recommender Systems (TORS), steering committee member of ACM Conference on Recommender Systems (RecSys), associate editor of International Journal of Human-Computer Studies (IJHCS), editorial board member of User Modeling and User-Adapted Interaction Journal (UMUAI), and associate editor of ACM Transactions on Interactive Intelligent Systems (TiiS). She has also served as program co-chair of IUI'26, general co-chair of ACM RecSys'23, program co-chair of ACM RecSys'20, and program co-chair of ACM UMAP'18.
Keynote #2
Title: Agentic Sensing for Wellbeing: Toward Trustworthy, Proactive AI That Reasons About Human Behavior, by Subigya Nepal
Abstract
Wellbeing may be the hardest place to get personalization right: the states we most want to detect are dynamic, affective, and often unspoken, and a wrong recommendation is costly. In this keynote, I argue the path forward pairs large language models with passive behavioral sensing (the continuous signals from phones and wearables) while taking evaluation as seriously as wellbeing demands. Drawing on recent work across student, emerging-adult, and clinical populations, I show how grounding LLMs in behavioral context can make personalized support more meaningful, why rigorous evaluation sometimes reveals that more AI is not more helpful, and how agentic systems that autonomously investigate behavioral data point toward proactive, just-in-time support that is adaptive, trustworthy, and accountable to real human outcomes.
About Subigya Nepal
Subigya Nepal is an Assistant Professor of Computer Science at the University of Virginia, where he directs the SONDER Lab (Sensing, Observing, aNd unDerstanding human ExpeRience). His research asks how the digital signals woven into everyday life, from phones and wearables to laptops and social media, can help us understand human behavior and mental health as it actually unfolds, and how AI can turn those signals into timely, context-aware support. He earned his PhD at Dartmouth College with Andrew Campbell, where he led some of the longest-running mobile-sensing studies of students, workers, and clinical populations, including a four-year study of college students that won a Distinguished Paper Award at UbiComp 2025. Before joining UVA, he was a postdoctoral fellow at Stanford's Institute for Human-Centered AI with Gabriella Harari, studying human-AI interaction for wellbeing.