RecTemp 2026

Time, Taste, and Language: Modeling Evolving User Preferences: from Temporal Dynamics to Large Language Models

Most recommender systems still assume that each user has a latent vector describing their tastes and preferences. It is generally acknowledged, however, that for many users this is false in a specific way: a long history is not necessarily repeated evidence about current preference, but a mixture of several past preference states. Aggregating across these preference states does not reduce variance; it potentially introduces bias. This talk briefly traces how the field has tried to address this fallacy. We review some of the key milestones in time-aware collaborative filtering, sequential and session-based models, architectures that make the short-term/long-term split explicit, and continuous-time models of repeat and periodic consumption. We argue that short-term versus long-term, however productive, is a two-state approximation of a richer process, and we offer a taxonomy of preference patterns. Most deployed systems model only a few of these patterns, leaving open potentially fruitful avenues for future research. The last part of the talk turns to large language models, briefly summarizing recent work and considering how they make natural language a viable user profile, one in which periodicity, seasonality and context become more expressible and inspectable. We close with recent results on LLM-driven temporal profiling and calibration under dynamic preferences.

BIO:
Dr. Bamshad Mobasher is a professor of Computer Science and the director of the Center for Web Intelligence at DePaul University College of Computing and Digital Media in Chicago. He is also director of the DePaul AI Institute, and interdisciplinary institute with a focus on the applications and ethical implication of AI technologies. He received his Ph.D. in Computer Science at Iowa State University in 1994. His general research areas include artificial intelligence and machine learning. In particular, he is considered one of the leading authorities in algorithmic personalization and recommender systems. He has published 5 books and over 300 scientific articles on topics related to the applications of AI and machine learning in user modeling, automatic personalization, and recommender systems. As the director of the Center for Web Intelligence, he directs research in these areas and regularly works with the industry on various joint projects. He has served in leadership positions of several prominent international conferences, including the ACM Conference on Recommender Systems, ACM SIGKDD Conference on Knowledge Discovery in Data, and the ACM Conference on User Modeling, Adaptation, and Personalization. Dr. Mobasher serves as an associate editor for the ACM Transactions on the Web, the ACM Transactions on Internet Technology and the ACM Transactions on Intelligent Interactive Systems, and the ACM Transactions on Recommender Systems. He has also served on the editorial boards of several other prominent computing journals, including User Modeling and User-Adapted Interaction, and the Journal of Web Semantics..