.png){kind=icon}
.png){kind=icon}
Return
A Multimodal Implicit Q-Learning Based Football Goalkeeper Performance Evaluating Model
.png){kind=icon}
PRE
AI
L
X
A
L
X
Z
DOI：10.1109/tbdata.2026.3688991.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Football is one of the most popular sports in the world, enjoying a vast and passionate following. With the continuous growth of the influence of the sport, football analysis techniques have evolved rapidly recently, becoming an important focus of research and innovation in academia. Although many existing approaches effectively assess outfield players, the unique characteristics of goalkeepers, such as long periods without ball possession and sparse event data, mean that evaluation of goalkeeper performance remains insufficiently explored. To address this gap, we propose Goalkeeper Defensive Value (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">GDV</i>), a goalkeeper performance evaluation model based on Implicit Q-Learning (IQL) with multimodal data fusion, specifically designed to assess goalkeeper defensive performance. By integrating high-frequency tracking data and event data into a multimodal representation, the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">GDV</i> captures the spatiotemporal dynamics of match scenarios and identifies patterns in goalkeeper performance. Three key modules are introduced to enhance IQL in the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">GDV</i>, including a modality-aware fusion module that provides enhanced spatial awareness, an action-aware modulation module that strengthens the tactical association between state and action, and a temporal encoding module that reveals global temporal dependencies of actions. Simulation results demonstrate that the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">GDV</i> effectively overcomes limitations of existing approaches in capturing off-ball goalkeeper behavior and evaluating contextual value of goalkeeper actions, providing a novel approach for goalkeeper performance assessment.
Keywords:
Goalkeeper performance evaluation
reinforcement learning
spatiotemporal modeling
implicit Q-learning
multimodal fusion
Journal
I
IF:
5.7
Papers:
834
Citations:
3.0K
Organization
Cited Papers