Conversational bandits combine clicks and comparisons for better recommendations
Conversational Generalized Linear Bandits with Hybrid Feedback
Machine Learning
Summary
Many recommendation systems ask users questions to better understand their preferences. Existing methods treat all feedback as the same, but people often rate items like movies with stars or clicks, while they compare words or features differently. The authors study a method that mixes these two feedback types — absolute item ratings and relative comparisons of key terms — in one system. They create an algorithm called HyConUCB that learns efficiently from both kinds of feedback and shows it leads to better recommendations in tests.
What this means in practice
- •For recommendation system engineers: Improve user preference elicitation by combining clicks and comparison queries for more accurate item recommendations.
- •For online marketing teams: Design interactive tools to engage customers by querying product features and collecting mixed feedback for targeted offers.
Authors
Yurui Zou, Qirun Zeng, Jinhang Zuo
Abstract
Conversational bandits improve recommendations by actively querying users about key-terms to elicit their item preferences. However, existing methods typically assume homogeneous feedback for items and key-terms, using either absolute rewards or pairwise comparisons. In practice, item feedback often takes the form of clicks or ratings, while comparisons are more natural for key-terms. To address this gap, we study conversational generalized linear bandits with hybrid feedback, combining absolute item rewards and relative key-term comparisons. We propose Conversational UCB with Hybrid Feedback (HyConUCB), which constructs unified confidence bounds from heterogeneous feedback and adaptively selects key-term comparisons by balancing information gain and directional coverage. Through a refined regret analysis that jointly exploits both feedback sources, we establish a regret bound that improves upon reward-only and adapted dueling-only benchmarks. Experiments on synthetic and real-world datasets validate the effectiveness of our algorithm.