Published on 10 June 2020
Researchers from the University of York have presented a ground-breaking data and technology platform, Weavr, to fans at the finals of ESL Birmingham Online, a tournament for professional esports players.
Developed at the University of York in collaboration with industry partners, Weavr delivers data-driven audience experiences across multiple realities, including virtual and augmented reality.
Powered by AI, the platform presents fans with interactive narratives and rich visualisations, providing real-time updates of the virtual arena, showcasing important performances and live statistics and a personalised compilation of on-demand highlights.
Athanasios Kokkinakis from the University of York comments, “The weekend was interesting. We successfully fielded machine learning models that in real-time converted complex data streams to spectator-friendly stories, but where you can also dive into the details. Along the way we have a fantastic opportunity to build a better understanding of how esports and sports analysis can be made more accessible and trustworthy.”
The online tournament involved the introduction of new features across multiple products included the Weavr mobile app, the Weavr virtual reality experience and the Weavr Twitch experience, all guided by University of York research.
These new features include state-of-the-art machine learning models that take the incredibly complex and high-frequency dataspace of real-time esports matches and find the trends, patterns and interesting events. Simon Demediuk, University of York, notes: “Esports are incredibly challenging from a data-perspective. We need to monitor thousands of variables in real time. With our new roll-out, Weavr is capable to automatically detect and calculate key athlete performances. We can also use live data to predict the outcome of matches.”
For esports enthusiasts who do not have time to catch a match, the work of Olu Olarewaju, University of York, provides automatically generated summaries, generated using Natural Language Processing and machine learning. “Our new algorithm intelligently detects key events across DOTA 2 matches, and uses analysis of real-life post-match commentaries by human presenters to create stories about matches. Because the system is intelligent, it can take into account your personal interests and give you match recaps that are personalised to your preferences,” comments Olu.
Personalisation is also a key element in the work of Justus Robertson, University of York. His system works in real-time during esports matches, to not only provide highlights of interesting events to viewers, but also to connect them into data-driven stories. Justus comments, “Our new rollout feature Story Threads. These are sequences of dynamically generated data-driven narrative arcs. The sequences are generated based on performance, in-game events, and historical data. They are then filtered and presented based on each user’s unique preferences.”
The new customisable Twitch extension brings the power of the Weavr data system to the world’s most popular games viewing platform providing DOTA 2 fans with real-time data at their fingertips.
Alan Pedrassoli Chitayat, a University of York alumni headhunted to the Weavr team even before graduating, is building models tackling the complex problem of vision in sports. Alan comments, “Vision is a form of information acquisition. It is critical in esports as well as in sports. If your team has vision in the game, they have good information. Games commonly have imperfect information built into the core tactics, and this is highlighted in games like Dota 2 where the position of the enemy team is not as easily available as it might be in a sport such as football.” Alan’s machine learning models demonstrate the value of vision and through these, the Weavr team is able to make certain predictions about DOTA 2 matches.
Moni Patra, University of York, highlights the way that the Weavr project connects data and storytelling to audiences: “Esports games are incredibly complex. From start to finish, the action is rapid and our models need to be able to adapt to a vast data space in real-time. A good example is the draft phase, where the teams select which heroes to play. Here our models need to take into account millions of possible combinations. I was very happy with the way our Bayesian models predicted which heroes teams would select. It makes for a great watching experience,” says Moni.
Florian Block, R&D Director, Weavr and Anders Drachen, Head of Analytics, Weavr, are part of the team that has built the largest university-based esports research environment in the world, where teams work across disciplines with the esports community. Anders Drachen notes, “At the University of York, we are fortunate to have world-class environments across multiple departments supporting innovation in esports. We are breaking new ground in how we use data to empower audiences to control their personal viewing experiences.”
Florian Block adds: “Being able to work with real esports fans in the context of large international esports events provides us with valuable insight on how AI-driven storytelling can add value to viewers and tournament organisers.”