Hierarchy2vec–Representation Learning for the Prediction of Hierarchical Teams’ Performance

Part of the Lubar Research Seminar Series

Speaker: Kang Zhao, University of Iowa

Teamwork is ubiquitous in today’s society and better predictions of team performance have important managerial implications for recruiting and personnel selection. Meanwhile, most teams feature internal hierarchies among team members, yet previous research on team performance predictions has not paid attention to such hierarchies. Thus, this study attempts to leverage hierarchical structures inside teams to better predict the performance of teams. The proposed hierarchy2vec model starts with node representation learning from historical collaboration networks with different types of collaboration ties. Then the model aggregates individual members’ representations into a team representation by following the team’s hierarchical structure and incorporating attention mechanisms. Experiments based on coaches in the National Football League demonstrate the effectiveness of the proposed approach.