Keynote #1

Title: On Human Preference Alignment for LLM-based Recommenders, by Xiang Wang

Abstract

Human-centric preference alignment has emerged as a frontier research direction in large language models, aiming to ensure that model outputs are aligned with human values and preferences. We apply preference alignment techniques to recommendation scenarios and propose the S-DPO framework, which effectively integrates preference alignment with recommendation-specific ranking objectives. The goal is to instill ranking information into the language model, enabling LLM-based recommenders to distinguish preferred items from negatives rather than focusing solely on positives. Moreover, we extend it with a self-improving mechanism.

About Xiang Wang

Xiang Wang is a Professor at the University of Science and Technology of China (USTC). His research interests include information recommendation and mining, large language models, and trustworthy AI. He has published over 70 papers in top international conferences and journals such as SIGIR, WWW, NeurIPS, ICLR, IEEE TPAMI, and ACM TOIS. His work has received more than 20,000 citations on Google Scholar, with an H-index of 56, and he has been recognized as a Highly Cited Researcher by Elsevier China. He has received the ICLR 2025 Outstanding Paper Award, two Frontier Science Awards at the International Congress of Basic Science, the ACM SIGIR Early Career Award and the Wu Wenjun AI Excellence Award in Natural Science in 2024, and was selected for the MIT Technology Review’s TR35 list.

Keynote #2

Title: From Recommendations to Interactions: Putting Users Back in the Loop, by Aixin Sun

Abstract

Recommender Systems (RecSys) have garnered significant attention from both industry and academia over the past decades. Considerable effort has been devoted to developing algorithms that generate more accurate recommendations by learning user preferences from past user-item interactions. However, while these historical interactions are typically collected from online RecSys platforms, the user’s role in the interaction process is often underexplored—particularly from the moment recommendations are presented to the point when user-item interactions are observed. In this talk, I will delve into the complexities of user-item interactions, including decision-making costs, multi-step engagements, and unobservable behaviors that can influence both model design and loss functions. Additionally, I will discuss the trade-offs between task specificity and model generalizability, emphasizing how well-defined task formulations provide the foundation for robust evaluation and effective solution development.

About Aixin Sun

Dr. Aixin Sun is an Associate Professor and Associate Dean (Undergraduate Education) at the College of Computing and Data Science (CCDS), Nanyang Technological University (NTU), Singapore. He received his B.A.Sc (First Class Honours) and Ph.D. from NTU in 2001 and 2004, respectively. His current research interests include information retrieval, recommender systems, and natural language processing. Dr. Sun has published over 200 papers, which have collectively received more than 22,000 citations on Google Scholar, with an h-index of 68. He serves as an associate editor for ACM Transactions on Information Systems (TOIS), ACM Transactions on Recommender Systems (TORS), ACM Transactions on Intelligent Systems and Technology (TIST), and Neurocomputing, and is also on the editorial board of the Journal of the Association for Information Science and Technology (JASIST). He has held various roles in academic conferences, including Search Track co-chair for WWW 2024, Doctoral Consortium co-chair for WSDM 2023, demonstration track co-chair for SIGIR 2020, ICDM 2018, and CIKM 2017, and program co-chair for AIRS 2019.