Facilitating Team Formation based on Users' Feedback: A Multi-Armed Bandit Approach
Team formation algorithms assist users in finding teams by leveraging users' characteristics, skills, and relationships. Despite recent advancements, most existing team formation algorithms do not consider user preferences when searching for optimal team combinations. Moreover, most of them do not consider when multiple users search for and choose others for their teams, which changes the number of possible team combinations. In this work, we formulate the team formation problem as a multi-armed bandit problem. Our approach sequentially selects teams from multiple pre-formed team combinations to find which ones are the most liked by users. The selection is based on contextual information about both the users and teams, while adapting its team selection strategy based on user feedback to maximize overall user satisfaction. The contributions of this work are twofold. First, we propose a novel application of a contextual bandit algorithm to the domain of team formation. Second, through extensive experiments on simulated data from different real-world datasets, we demonstrate how our proposed algorithm learns from user feedback significantly enhances the relevance of team recommendations compared to benchmark algorithms.