TLDR: This research introduces a framework to transform football data into object-centric event logs, enhancing traditional process mining by considering multiple interacting objects like players, teams, and the ball, along with spatial context. This approach provides a more comprehensive understanding of game dynamics, including off-ball actions and player movements, which is crucial for analyzing complex tactical behaviors and improving football analytics.
Football, often dubbed ‘the beautiful game,’ is a complex tapestry of individual actions and team strategies. Analyzing this complexity to understand game performance and tactical decisions has long been a goal for coaches, analysts, and researchers. Traditional methods of process analysis, often relying on ‘event logs’ that focus on a single aspect, have struggled to capture the full picture of a football match.
A new research paper, titled “Transforming Football Data into Object-centric Event Logs with Spatial Context Information,” introduces a groundbreaking framework that promises to revolutionize how we understand football data. Authored by Vito Chan, Lennart Ebert, Paul-Julius Hillmann, Christoffer Rubensson, Stephan A. Fahrenkrog-Petersen, and Jan Mendling, this work addresses the limitations of conventional analysis by adopting an ‘object-centric’ approach.
The Challenge with Traditional Analysis
In conventional process mining, an ‘event log’ typically tracks events related to a single ‘case’ – for example, only actions involving the ball. While useful, this approach often overlooks crucial aspects of the game, such as the movements and behaviors of players who are not directly interacting with the ball. This can lead to an incomplete understanding of team tactics and individual player contributions, especially for ‘idle’ players who might be strategically positioned.
Introducing Object-Centric Process Mining
The paper proposes using object-centric event logs (OCELs), which allow events to be associated with multiple ‘objects’ simultaneously. Imagine tracking not just the ball, but also every player, each team, and even specific grid positions on the field, all at once. This multi-faceted view provides a much richer and more realistic representation of the game’s dynamics.
The researchers developed a framework that defines three main event classes for football data:
- Game-based events: General events like fouls or cards.
- Ball events: Actions directly involving the ball, such as passes, shots, or free kicks.
- Position-based events: Events related to player movements and their locations on the field.
Crucially, the framework also identifies six key object types:
- Match: The entire game itself.
- Team: Each competing side.
- Player: Individual athletes on the field.
- Possession: A period when one team controls the ball.
- Grid position: Specific areas on the football field, dividing it into a simplified spatial map.
- Ball: The central object of the game.
By linking events to these multiple objects, the framework can answer complex questions. For instance, instead of just knowing where a shot occurred, you can also know which player took it, which team they belonged to, during which possession it happened, and even the movements of other players leading up to that shot.
Putting the Framework to the Test
To validate their framework, the authors generated an object-centric event log using real-world football data from Metrica Sports. They then compared analyses from a single-object perspective (focusing only on the ball) with a multi-object perspective (including ball, possession, and player objects).
The results were striking. While a single-object view might show a sequence of passes leading to a goal, the multi-object view revealed much more. It highlighted additional activities like “Player changes position,” showing how players moved between other actions. When visualized on a spatial map, the multi-object view provided insights into individual player movements and overall team shape during a possession, demonstrating how players moved to support an attack, even if they weren’t directly touching the ball.
For example, the analysis could show that while one player moved only slightly, another player covered significant ground across the field, highlighting their involvement in offensive plays. This level of detail is invaluable for understanding tactical decisions and player performance beyond just on-ball actions.
Also Read:
- Navigating Data Diversity: A Survey of Transformation Techniques for AI
- How AI Learns to Think Like Humans in Novel Contexts
A More Complete Picture of the Game
This research marks a significant step forward in football analytics. By transforming raw football data into object-centric event logs with spatial context, the framework provides a more comprehensive and realistic view of a football game. It enables a deeper understanding of team tactics and individual player behaviors, including the often-overlooked contributions of players not directly involved in ball actions.
While football games are inherently complex and variable, this object-centric approach offers a powerful tool for analysis. Future work will focus on techniques to handle this variability and explore its application to other team sports. You can read the full research paper here.
