A Temporal Graph Model to Study the Dynamics of Collective Behavior and Performance in Team Sports: An Application to Basketball
Quentin Bourgeais, Eric Sanlaville, Rodolphe Charrier, Ludovic Seifert
TL;DR
The paper introduces a Temporal Passing Network Model (TPNM) that embeds the temporal dimension into basketball passing interactions by modeling possessions as sequences of graphlets in rolling time windows. It defines three entropy-based metrics ($\mathrm{SE}$, $\mathrm{TE}$, $\mathrm{RTE}$) to quantify network complexity and degeneracy, and tests two hypotheses: higher entropy correlates with better performance, and the current relative score acts as a constraint shaping entropy. Using a 2019 FIBA World Cup dataset, the authors show a moderate, significant positive link between entropy and final score, and reveal how relative score contexts differentially influence network complexity across teams, yielding both common and unique team signatures. The TPNM provides a flexible, possession-level, temporally-aware framework that can be extended with additional data dimensions (spatial, opponent interactions) and applied to other social networks, while acknowledging limitations in data size and observational nature. Overall, the work contributes to understanding how complex, adaptable interaction patterns under constraints relate to performance in team sports and offers a pathway for cross-sport applicability and interventions.
Abstract
In this study, a temporal graph model is designed to model the behavior of collective sports teams based on the networks of player interactions. The main motivation for the model is to integrate the temporal dimension into the analysis of players' passing networks in order to gain deeper insights into the dynamics of system behavior, particularly how a system exploits the degeneracy property to self-regulate. First, the temporal graph model and the entropy measures used to assess the complexity of the dynamics of the network structure are introduced and illustrated. Second, an experiment using basketball data is conducted to investigate the relationship between the complexity level and team performance. This is accomplished by examining the correlations between the entropy measures in a team's behavior and the team's final performance, as well as the link between the relative score compared to that of the opponent and the entropy in the team's behavior. Results indicate positive correlations between entropy measures and final team performance, and threshold values of relative score associated with changes in team behavior -- thereby revealing common and unique team signatures. From a complexity science perspective, the model proves useful for identifying key performance factors in team sports and for studying the effects of given constraints on the exploitation of degeneracy to organize team behavior through various network structures. Future research can easily extend the model and apply it to other types of social networks.