Efficient Explorative Key-Term Selection Strategies for Conversational Contextual Bandits

Conversational contextual bandits elicit user preferences by occasionally querying for explicit feedback on key-terms to accelerate learning. However, there are aspects of existing approaches which limit their performance. First, information gained from key-term-level conversations and arm-level recommendations is not appropriately incorporated speed up Second, it important ask explorative quickly the user's potential interests in various domains convergence preference estimation, has never been considered works. To tackle these issues, we first propose ``ConLinUCB", a general framework conversational with better incorporation, combining estimate one step at each time. Based this framework, further design two bandit algorithms key-term selection strategies, ConLinUCB-BS ConLinUCB-MCR. We prove tighter regret upper bounds our proposed algorithms. Particularly, achieves bound than previous result. Extensive experiments synthetic real-world data show significant advantages learning accuracy (up 54% improvement) computational efficiency 72% improvement), compared classic ConUCB algorithm, showing benefit recommender systems.