Sambagraph models soccer tactics using player and ball interaction graphs
SambaGraph: Action-Reaction Spatio-Temporal Graphs for Soccer Tactical Response Modeling
Machine Learning
Summary
Soccer tactics involve back-and-forth actions between teams, where what one team does changes how the other responds. The authors created SambaGraph, a big set of graph-based data showing player and ball interactions during World Cup games to study these responses. They tested how well computers can classify the defensive reactions, find good defensive moves, and explain tactics. While some machine learning methods worked well for identifying responses, language models gave better explanations but worse predictions. SambaGraph offers a new benchmark for studying soccer strategy with graphs.
What this means in practice
- •For soccer coaching teams: Identify common defensive responses to opponent attacks from real match data to inform tactical training.
- •For sports analytics companies: Develop tools that retrieve successful defensive setups matching an observed attack pattern for strategic analysis.$Commercial implications: Enables advanced AI-powered tactical analysis software targeted at professional sports teams and broadcasters.
Authors
Abel A. Reyes-Angulo, Henry O. Velesaca, Steven Araujo
Abstract
Soccer tactics are interactive: an attacking action changes the opponent's defensive problem, and the observed response depends on the multi-agent match state. We introduce SambaGraph, an action--reaction spatio-temporal graph dataset and benchmark for soccer tactical response modeling. From tracking and event data for all 64 matches of the 2022 FIFA World Cup, we curate 4,070 action-centered episodes represented as temporally aligned 23-node player--ball graph sequences with attack/defense views, response labels, and 26,270 split-safe attack--defense pairs. We study three questions: whether observed responses can be classified from graph episodes, whether successful defenses can be retrieved for a query attack, and whether graph-derived summaries support grounded LLM reasoning. A compact signature MLP obtains $0.796\pm0.007$ macro-F1 for response classification, while a fused graph--signature dual encoder reaches $0.471\pm0.029$ Hit@5 and $0.655\pm0.051$ Hit@10 for full-bank defensive retrieval. Hard negatives maximize pair discrimination but not retrieval quality. Local LLMs underperform supervised encoders for direct classification and do not improve over a strong original order in eight-candidate reranking, but they provide grounded tactical rationales. These results position SambaGraph as a reproducible benchmark for graph-based soccer strategy-response research. Code and dataset are available at: https://github.com/areyesan/SambaGraph.