Gabriel Anzer, Pascal Bauer, Ulf Brefeld, Dennis Fassmeyer
Overlapping runs are a widely used group-tactical pattern in soccer. By combining variational autoencoder with a graph neural network representation of positional data, we are able to detect overlapping runs using only a very limited amount of hand-labeled data. Based on this detection, we show practical applications using data of the German national team during the European Championship 2021. Using the same methodology, we outperform state of the art approaches on the prediction of player trajectories using a publicly available Basketball dataset.