In this study, we define a new way of representing football player roles based on passing and receiving interactions. We develop a definition of player roles consisting of a linear combination of 12 common and interpretable passing/receiving patterns. Linear combinations are derived from the decomposition of players` pitch passing and receiving networks using non-negative matrix factorization (NMF). Our model shows that 43% of the 1491 players studied in this paper had a maximum weight of less than 50% in each of the 12 common passing/receiving patterns. This suggests that a substantial percentage of players do not follow the specific passing/receiving patterns typically associated with their conventional role. The model also reveals the underlying differences in passing/receiving patterns amongst players who hold the same conventional role. It shows the intricacies of player patterns optimally when tasked with analyzing the most complex conventional roles such as midfielders, wingers, and forwards. Lastly, we show that the combinations of the 12 common passing/receiving patterns can be used as a footprint to find players with similar passing/receiving styles. For instance, our model found that Shaqiri and Fabinho had the highest similarity in passing/receiving styles to Oxlade-Chamberlain and Henderson. This is consistent with Liverpool FC`s transfers of Shaqiri and Fabinho to replace Oxlade-Chamberlain and Henderson`s positions respectively in the summer of 2018.
© Copyright 2022 Machine Learning and Data Mining for Sports Analytics. MLSA 2021. Communications in Computer and Information Science. Veröffentlicht von Springer. Alle Rechte vorbehalten.