Cognitive

IBM Player Spotlight Built with IBM Watson

Share this post:

The Emerging Technology team within IBM Research UK, have been working with FOX Sports to enhance their viewer’s experience of watching ‘The Beautiful Game’.

Example broadcast content shown on Fox Sports, produced by the IBM Player Spotlight Built with IBM Watson tool.

The IBM Player Spotlight Built with IBM Watson is an AI-powered tool built for the FOX Sports’ studio team to access complex statistical analyses using an easy-to-use conversational interface. With the analysis clearly presented, they are then able to use these insights, gathered for both individual players and teams in their on-air broadcasts.

This year’s FIFA Women’s World Cup is the 8th in the tournament’s 18 year history. It is being hosted in France with 24 teams in the competition, culminating in the final 2 teams playing on 7th July at Stade de Lyon. Previous winners include the United States (3 wins), Germany (2), Japan (1) and Norway (1), but who will it be this time around?

What did you want to analyse today?

To enable users to quickly and easily view the data they wish to analyse, the solution includes a trained instance of IBM Watson Assistant. Watson is able to take the input from the user and understand the football-specific terms included, for example, team names, player names, locations on the pitch and events in the game.

Demonstration of the complex football terminology that can be used, here shown by the playlist returned when asking Watson “Show every cross for England on the right wing in the second half”

Using Watson in this way, we are able to gather the meaning/intent of the question the user is asking. Watson replies with analysis, matching video footage and other data points, then allowing the user to uncover insights into the sport that would have previously taken many hours of detailed video and statistical analysis to find.

Extrapolating the data to capture playing styles

The XML based Opta data that is fed into our platform, details information about every on-ball event in every game across the competition. However, this data in isolation isn’t useful from an analysis point of view. For example, how can I find out about England’s attacking style? Something we could ask Watson with “Show me England’s plays in the final third’, but how can we expect it to understand the meaning of that? What is an England play? What is the “final third”?

An example broadcast graphic that demonstrates a comparison of playing styles between Sweden and the Netherlands. The model categorises each . play the team undertakes into one of these categories, it’s then possible to ask Watson “Compare the Netherlands and Sweden” and get this output.

To be able to accurately model the passages of play in a game of football, we use a machine learning model, trained in IBM Watson Studio. This machine learning model is trained with historical footage from games as well as custom statistics that we have generated in-house from our own understanding of the game. As such, we are able to group together events from the Opta dataset to form a ‘play’. These plays are not only limited to a particular event, for example a goal, but will show the context of the build-up to that goal, how the play started and resulted with the ball going into the net..

Having these plays in place, we are then able to apply another model to categorise the style for a particular play. For example, is this showing us wide play? Is this “route one” play? Now that we have this high-level classification for a play, as well as all the events that make up that play, we have more context to work with. This enables us to present the most useful information to the user in each scenario, be it as video clips of a particular play or breakdowns of areas of play on the pitch, areas where a team are more vulnerable.  This holistic view of all the data in one place provides huge opportunities for analysis and insight by the users.

So, what kind of insights can we see?

There really are a lot of them!  And because of the way IBM Watson Assistant responds to our questions, we can ask for things in a number of ways. Let’s explain with a number of example questions, and a breakdown of the kind of analysis that can be performed.

“SHOW ALL SHOTS FOR THE USA IN THE SECOND HALF”

Screenshot showing a sample of the stats returned by Watson detailing high-level insights based on the question asked.

The first image here shows the Playlist of all 43 plays which we have in the system (at time of writing) that are a shot by the USA in the second half of their World Cup matches. Clicking on these provides an animation of the passage of play between players that led up to the shot, including hover details of each player and any key data about that event.

The Stats tab shows a more detailed breakdown of these plays, with the bar graph clearly showing that we’re only looking at second half plays. Interestingly, this tab is also telling us that of all the second half shots the USA have had in the tournament so far, 30% of them (13) have resulted in a goal. More analysis shown on this screen (not visible in this screenshot) gives a breakdown of the types of plays that have resulted in these shots. In the case of the USA, we see a possible aversion to long ball play and more focus on wider or more direct play to get a shot.

“SHOW THE PLAYERS THAT ARE INVOLVED THE MOST FOR ENGLAND”

Response from Watson detailing that Lucy Bronze is the most involved player for England.

From this data we can see the “play depth chart” for Lucy Bronze on the right (and every other player if we scrolled down). This chart shows where during the play Lucy was involved. The left of the chart is defined as the start of the play, possibly a throw-in or interception and the right of the chart is defined as the end of a play, likely to be the ball going out of play or a goal. For Bronze, the left and right sides are high, with a dip in the middle. Knowing that Lucy Bronze is a defender, the fact that she’s been involved at the start of plays is hardly surprising, but seeing her involvement in plays at the end too shows that she’s an important attacking asset for England and potentially one for other teams to watch.

Further analysis of this involvement data also showed us that of all England players, Nikita Parris is far more likely to run at people and take them on than any other player. It is also showed that the key players for England in terms of ball distribution are Lucy Bronze and Stephanie Houghton, so neutralising their threat could make England’s possession game more difficult.

“WHAT IS THE PASS NETWORK FOR FRANCE SHOTS?”

Screenshots showing the interactive network chart that details the most common passes between French players leading to a shot. In this example, the data has been filtered to emphasise passes made to/from Amel Majri.

The image below shows the most common passes that lead to a shot for the French team. Understanding this is often key to understanding how a team attacks, and especially having this understanding as an opposition would be key to minimising the risk of conceding shots and goals. The right-hand side of this screen shows the “strongest links” in the team and the entire dataset is interactive and filterable depending on where the analyst/producer wants to deepdive.

“SHOW THE PASS HEATMAP FOR LINDSEY HORAN”

The query here is showing us the ‘pass heatmap’ for Lindsey Horan, who primarily plays in a midfield role for the USA.

Lindsey Horan’s Pass Heatmaps. (top-left) Horan’s position for passes she made, (top-right) Where Horan’s passes ended, (bottom-left) Passers position when passing to Horan, (bottom-right) Horan’s position when receiving a pass.

The top two images show her position when making a pass (top left) and the position of the person receiving the ball from her (top right). We can see from this that more often than not her starting position to make a pass is in the middle third of the pitch, and that her ball distribution tends to cover a large area of the attacking half of the pitch, with a concentration out on the left-wing (attacking direction is from left to right).

Conversely, the bottom two images are showing the positions of all players passing to Lindsey Horan (bottom left) and her position as she receives the pass (bottom-right). Here we can see that the passes to Horan are from all over the middle third of the pitch, but that Horan herself tend to receive the ball more often around the centre circle. These graphics are also interactive, allowing the user to hover over the heatmap and see the passes going out of or into a particular section in any of the charts. This provides much more detail about where a particular player is targeting their passes when they get into a particular area of the pitch.

Broadcast Live On-Air

FOX Sports, at the point of publishing this article, have shown content from the platform on-air 8 times throughout the knockout stages, showcasing a range of insights and graphics. Each time, the insights are commentated on and then Fox have tweeted out short video segments to accompany the on-air broadcast.

Tweets by @FOXSoccer sharing the insights provided by IBM Watson through our platform.

Whilst the videos aren’t available to play in the UK, it’s still very rewarding to see our work being shared in this way. They’ve been generating discussion online and the platform has opened a new angle for producers and analysts to explore complex performance data themselves, without a steep learning curve. They can get answers to their questions in seconds and clearly present interesting information on-air to their audiences.

 

Master Inventor & Emerging Technology Specialist

Liz Maple

Senior Inventor & Emerging Technology Specialist

More Cognitive stories

Playing with the Bloch Sphere

How we made a game of learning Qubits . In this article we will briefly show the tool we built while learning quantum computing; alas, this article is not about teaching these concepts. We will be covering these in a series of articles soon.     At the start of our journey we were learning […]

Continue reading

Emerging Technology supports Project Owl within IBM Corporate Service Corps Deployment

In March 2019, Dan Cunnington from Emerging Technology was deployed to Puerto Rico to help facilitate the field testing of the winning Call for Code solution – Project Owl. As part of the grand prize, Project Owl won a field testing opportunity supported by IBM Corporate Service Corps to field test their winning solution on […]

Continue reading

Helping Disabled Passengers Travel with Confidence

Introduction Disabled or disadvantaged passengers have every right to fear travelling with multiple high profile cases of unfair and appalling levels of service evident in the media, particularly for rail passengers. Our team have been tackling this problem recently in collaboration with the IBM Travel and Transport team and 50% funding through Innovate UK’s call […]

Continue reading