Transformer-Based Neural Marked Spatio Temporal Point Process Model for Football Match Events Analysis

The NMSTPP model architecture:

The code and dataset for the proposed model are contained in the following files.

1. dataset_create.py creates the csv dataset from the json file retrived from the WyScout Open Access Dataset referencing the code for the Seq2event model and the corresponding heatmaps for the grouped actions (pass, dribble, cross, shot, and possession end).

2. train_valid_test_split.py creates the other continouses features and split the csv dataset created in step 1 into train/valid/test set according to the 0.8/0.1/0.1 ratio.

3. NMSTPP.py is the code for training, validating, and testing the NMSTPP model referencing the code for the Seq2event model.

4. forecast.py is the code for forecasting the entire dataset with NMSTPP model.

Introduction

With recently available football match event data that record the detail of the football match, analysts and researchers have a great opportunity to develop new performance metrics, gain insight, and evaluate key performance. However, most sports sequential events modeling methods and performance metrics approaches could be insufficient for dealing with such large-scale sequential data, necessitating a more complex model and holistic performance metric.

To this end, we propose the Transformer-Based Neural Marked Spatio Temporal Point Process (NMSTPP) model for football event data based on the neural temporal point processes (NTPP) framework. In the experiments, our model outperformed the prediction performance of the baseline models. Furthermore, we have proposed the holistic possession utilization score (HPUS) metric for a more comprehensive football possession analysis. For verification, we examined the relationship with football teams' final ranking, average goal score, and average xG over a season, the average HPUS showed significant correlations regardless of not using details shot or goal information.

Reference

For technical details and full experimental results, please check the paper. Please consider citing our work if you find it helpful to yours:

@article{yeung2023transformer,
  title={Transformer-Based Neural Marked Spatio Temporal Point Process Model for Football Match Events Analysis},
  author={Yeung, Calvin CK and Sit, Tony and Fujii, Keisuke},
  journal={arXiv preprint arXiv:2302.09276},
  year={2023}
}