The 2023 StatsBomb Conference was held at Wembley Stadium just over a month ago. As part of the event, we invited the winners of our Research Competition to showcase their work to an audience of industry experts and professionals.
We received numerous submissions, and the quality of the proposals was incredibly high, making the selection process quite challenging. In order to recognise all of the best entries, we introduced a new addition to the conference this year: research posters. These displays provided a platform for researchers to showcase and discuss their work with the conference attendees.
Additionally, we created several posters displaying the latest research from our very own data science and analysis teams. We are excited to offer you a closer look at this work, covering a wide range of topics such as team tactics, large language models, player evaluation, and more.
The individual posters can be downloaded below, simply by clicking on the image.
An Events and 360 Data-Driven Approach for Extracting Team Tactics and Evaluating Performance in Football
By Calvin Yeung and Rory Bunker
In the context of this project, team tactics refers to how individual players adjust their positioning when they come up against opposition to create a successful attack. These tactics can be observed during periods of possession.
TacticalGPT: Uncovering the Potential of LLMs for Predicting Tactical Decisions in Professional Football
By Matthew Caron, M.Sc., Prof. Dr. Oliver Müller
This study explored the capability of LLMs (large language models) to deliver tactical insights for on-ball and off-ball situations through natural language.
A New Framework For Skill-Based Metrics
By Dinesh Vatvani, Will Thomson, Scott Johnson
Exploring skill-based metrics is one of the primary projects the StatsBomb data science team have been researching over the past year. The conference poster was the first public demonstration of that work, with an introduction to the methodology, findings and results of using a skill-based approach.
Scott Johnson, Senior Analyst, StatsBomb:
"The poster talks about our new skill framework which aims to give better estimates of player skills over time. It takes into account all the data we have for a player instead of relying on data from just one league-season as is typical in the field. We applied this framework to shooting and goalkeeper shot-stopping which yields some interesting results.
The benefit of the framework is that we can apply it to more skills in the future such as passing, dribbling etc. The research was a team effort by the Data Science team here at StatsBomb and it was great to put together a poster for the conference so that the community could engage with it."
What can xPass tell us about team and player style?
By Lily Wood-Blake & Jaymes Monte
This study used StatsBomb's new xPass model, released in October, to reveal insights into team and player style and ability. xPass estimates the success of an attempted pass, taking into account various factors that could impact its completion.
Jaymes Monte, Senior Analyst, StatsBomb:
"Working with our new xPass model gives the opportunity to add extra layers of context to a variety of different analyses - both from a performance perspective and in the recruitment environment.
The topics explored in this poster only started to scratch the surface of what is possible, hopefully they can inspire others to take the ideas and concepts further."
Data Drift Tracking
By Duncan Hunter
We recently implemented a system into our world class QA process to detact any changes to the data we collect. Duncan explains why we have done this, and how we assess the need for corrective measures.
Duncan Hunter, QA Data Scientist, StatsBomb:
"It was great to be able to showcase some of the behind the scenes work that I've been doing, and presenting a poster at the conference was a great way to discuss its impact with researchers.
The system we've implemented to track potential data-drift is now a key part of our data collection and quality assurance process, highlighting potential inaccuracies or changes in our data at the earliest opportunity and, ultimately, maximising its integrity."
If you are interested in Football Analytics, you can enrol in one of our online courses today.