An Interactive Recommender System with Incremental Utility Learning Event

Time:

11:00 - 12:00

Date:

2026-10-29 10:59:00

Venue:

Ketley Room B54/4001

Book this Event

Event details

Personalized recommendation requires learning individual user preferences from limited interaction data and using the learned preferences to guide subsequent recommendations. In this talk, we present an interactive recommendation framework that integrates preference elicitation, incremental utility learning, and personalized recommendation into a unified process.

The framework extracts pairwise preference information from users' ordinal click sequences during ordinary webpage browsing and incrementally updates a user-specific utility function as new feedback becomes available. The learned utility is then used to generate subsequent recommendation lists, balancing the exploitation of current preference knowledge with exploration for additional information. We investigate two utility representations within this framework. The first considers a linear multi-attribute utility function that characterizes users' preferences over deterministic product attributes. The second adopts a nonparametric von Neumann–Morgenstern (VNM) utility function to capture users' risk preferences when product outcomes are uncertain. We develop optimization-based incremental learning methods for both representations and employ exploration–exploitation strategies, including epsilon-greedy and entropy-regularized sampling, to support adaptive recommendation.

Numerical experiments in financial product recommendation demonstrate the effectiveness of the proposed approaches in learning user preferences and improving recommendation performance.

About the Speaker

Jiaxin Wei is a Ph.D. student in the School of Mathematics and Statistics at Xi’an Jiaotong University, China, and is currently a visiting researcher at the University of Southampton, hosted by Prof. Chao Zheng. Her research focuses on stochastic optimization, preference elicitation and personalized recommendation systems. She previously conducted research at The Chinese University of Hong Kong under the supervision of Prof. Huifu Xu. Her work has been published in Annals of Operations Research and Financial Innovation, with a focus on portfolio optimization under uncertainty and personalized fund recommendation.