Hierarchy2vec: Representation Learning in Hierarchical Collaboration Networks for Team Performance Prediction

Published in INFORMS Data Science Workshop, 2021

Sulyun Lee and Kang Zhao

This paper was nominated for “best student paper.” Certificate

Abstract:
In collaboration networks, synergies among team members play an important role in team performance. At the same time, many teams feature hierarchical structures, where some team members report to another with a higher level position. With members having different responsibilities or playing different roles in a team, collaboration patterns between team members organized in a hierarchy can have important implications for the team’s performance. Focusing on the hierarchical nature of collaboration networks, this work is the first attempt to learn vector representations of teams based on individual team members’ characteristics and how they are organized into a hierarchical team. The proposed hierarchy2vec model first learns node embeddings via hierarchically biased walks and then aggregates such node embeddings in a hierarchical way to generate the team’s vector representations. Experiments on a real-world dataset for coaches in the National Football League (NFL) reveal that the proposed model can achieve better results in team performance predictions.

Paper link: PDF

Recommended citation:
Sulyun Lee and Kang Zhao. Hierarchy2vec - Representation Learning in Hierarchical Collaboration Networks for Team Performance Prediction. INFORMS Data Science Workshop, 2021.