Which team is the best in the league? How does my team fare with respect to the rest of the league? These are questions that every sports fan is interested in knowing the answers to. In other cases, such as in college sports, knowing the answer to these questions is crucial for shaping the picture of specific contests. In professional sports, sports networks provide power rankings regularly - typically every week or month depending on the season length of the league - based on their experts opinion. In this work we propose an alternative, objective and network-based way of ranking sports teams. In brief, our method is based on analyzing a directed network formed between the teams of the corresponding leagues that captures their win-lose relationships. Using data from the National Football League and the National Basketball Association, we show that even simple network theory metrics (e.g., Page Rank) can provide a ranking that has the same accuracy in predicting winners of upcoming match-ups as more complicated systems (e.g., Cortana). We further explore the impact of the network structure on the prediction accuracy and we show that the cycles in the network are significantly correlated with the performance. We finally propose an advanced ranking technique based on tensor decomposition
© Copyright 2016 Proceedings of the KDD-16 Workshop on Large-Scale Sports Analytics. Veröffentlicht von Eigenverlag. Alle Rechte vorbehalten.