The Societal Potential of Esports Analytics

Research in esports has really hit the ground running in the past decade. In parallel with the rise in popularity of esports worldwide, research covering a wide variety of fields from law, physiology, health, marketing, strategy, team behaviour, and more, has emerged. Professional esports teams can now have their own groups of researchers analysing data, strategies, training methods and how to best interact with fans.

Given the incredible pace of innovation in esports – and in games in general – there is also an enormous amount of work being done on developing new technologies for broadcasting, data capture, VR/AR-enhanced coverage, AI-enhanced production tools, and so on. This is not even starting to dig into the actual game AI work that uses esports as a basis due to the incredible complexity of these games but also the constrained nature of these games, which makes them great as experimental testbeds.

But the research work in esports really started with the data. Since the first Day of the Ancients mod to Warcraft 3, data have been at the centre of esports. Possibly with the exception of MMORPGs, esports were the first games to really start sharing data with the public and third parties in a major, and consistent, way.

This made esports like Starcraft an obvious starting point for some of the first game AI research, and a key baseline for the sister domain of game analytics. With only a few exceptions, game analytics cut its teeth on esports data, because back in the early days that is what we had available – and to a substantial extent, esports data is still the inroad for many students and early career researchers who are interested in how we play games, irrespective of their scientific domain.

Now, a key question to ask is, why is this research important? What does it do for us?

Most domains of esports work explain themselves, and it is immediately obvious why esports is a great case study in those domains, from which we can draw knowledge applied in a much broader context.

For example, we do AI work in esports because we want to build better, more engaging games and better artificial agents, train new AI algorithms, and use that knowledge outside of esports also, e.g. in AI-based game testing or outside of games entirely. We do legal research to ensure athletes and other stakeholders are treated fairly. We do physiological research to find the best training regimes and protect the players.

We use Neurological methods to explore how we as humans learn new skills. We use Psychology in esports to better understand human communication and teamwork. We do accessibility research to ensure as many people as possible can participate in the esports experience and community.

But why do we do esports analytics? What benefits does this bring to the world? What real-life problems does this solve?

Beyond the professional players, analytics powers the broadcasting of esports, in a way that far surpasses what we see in more traditional sports broadcasting. Analytics is used to tell individualized stories to audiences, to make the complex accessible and thus opening up the sport to broader audiences than previously possible. Along the way, we are learning how to tell stories with data in traditional sports, and to make sports broadcasting more engaging.

Analytics is also used to translate the enormous amounts of esports data into formats where they can be adopted by the hundreds of millions of players, streamers and third-party service providers worldwide who use data and analytics to inform, entertain or improve their own performance. Or just engage with their favourite pro team.

Perhaps the most significant contribution of esports analytics – and here I am specifically talking not only about academic work, but the enormous community of hobbyist and professional analysts who translate data into insights, stats websites and storytelling – is to deliver the tools needed for hundreds of millions of children and young adults all over the world to engage with data. To learn data literacy. To deliver data about something they are intrinsically motivated to learn about, and at least the foundational tools for doing something with that curiosity. It is a grand experiment in science education and something that, despite its worldwide impact, has gone virtually unrecognized.

Data literacy is a key 21st-century skill, something that is already hugely important to function in modern society and will become vital in the future. The promise of teaching our children about data and how to use – and not use – and reflect critically on – data, is worth researching, supporting, and preserving.

Recognizing the wider societal impact of esports analytics is only something that has happened recently, thanks to pioneering researchers, creators and professionals like Florian Block and James Dean, Pyrion Flax and many, many others.

Going back to academic research, what is it we have been focusing on in esports analytics?

Thanks to the recent review by Lincoln Costa, we have a pretty good understanding of the state of the art of knowledge in the domain and what the flagship areas are.

To be honest, the knowledge base is perhaps more characterized by gaps in our knowledge than actual knowledge, but games research and esports research in particular, are young domains so this is what we would expect.

There are hundreds of papers on or utilizing esports analytics, but the flagship areas are clear:

  1. Prediction
  2. AI: human-centred AI, NPC/AI agent enhancement, etc.
  3. Recommendation and Training
  4. Toxicity
  5. Teamwork
  6. Game design
  7. Audience experiences
  8. Education

Let us start with Prediction. Predicting the outcome of esports matches is by far the dominating component of esports analytics. Prediction here focuses on trying to figure out what might happen in the future – whether within a given match, in future matches, or anything else really. For example, which hero/champion a professional team might pick or ban next during the pre-match phases. Predicting the value of talent is also a consideration, similar to traditional sports analytics. Most prediction work in esports analytics has focused on match prediction. This is a nicely constrained task which can be solved at a variety of complexity levels, making it deal as a student project or as the focus for an entire research team.

There is a somewhat unclear relationship between academic research on this topic and the billion-dollar esports betting industry. We do not know much about how knowledge migrates across from academia to this sector, but it means there are some fundamental ethical considerations we need to take on board when engaging in this kind of research.

Back in 2017 we decided to take a different path than previous research in the area. Our goal was not been to train the most accurate model but to train the model that would be as useful as possible for esports broadcasting – for audiences. To begin with, this meant building models that could operate in real-time, and give you a status of the match while watching.

As mentioned, esports are complex games, and for novice viewers, it can be hard to understand what is going on in a match. We have focused our efforts on trying to use prediction, and other machine learning tools, to make these games more accessible and integrating our models in actual broadcasting tools, applications and studio-based recording environments.

Since our first real-time model was finalized, we have built a lot of other similar real-time frameworks making predictions of various kinds, for example, the famous “death predictor” for multi-player arena games like DOTA 2 developed by James Walker and colleagues. These deep learning models predict who the next character to be killed will be, just a few seconds into the future, which is just enough time for broadcast teams to move the in-game camera to where the action will be. These micro-prediction models stand as the most complex machine-learning models in esports today and have immediate implications for traditional sports broadcasting.

Along the way, prediction work has also helped solve problems of data and athlete representation which are being ported to analytics in traditional sports, for example, Simon Demediuk’s work on building performance measures specific to the position of individual players, enables much more precise evaluations of performance at the individual level.

We are increasingly seeing traditional sports analytics employ esports data to experiment, build and test models, thanks to the massive amounts of data available, when compared to traditional sports.

Moving on to AI-based work, esports has been used as a vehicle for a lot of game AI work, to a degree where it is hard to know where analytics stop and AI begin, but in essence, we have adopted analytics to try and improve interfaces in esports, improve AI-driven agents, AI-based game testing, and how to adapt the game to the individual player, e.g. through dynamic difficulty adjustment. There is a lot more to be said here, this is a huge domain across industry and academia, but I am not an expert on game AI and will refrain from embarrassing myself further.

Recommendation systems are another popular topic and form the foundation for a lot of analytics work aimed at creating training tools for players or information systems for coaches and analysts. Recommendation can range from simple tools that recommend strategies or weapon layouts to players, to coaching systems that attempt to learn the player´s playstyle and provide suggestions on how to improve performance. Given the complexity of esports games, this is a challenging exercise but also highly rewarding. If we can build recommendation tools that work across the complexity of esports teams and matches, we can do the same thing in other contexts, for example, education, organisational teams, and more.

Toxicity research has become an important topic in online interaction in general, and in the context of esports forms a substantial challenge. Here work is focused on mining chat logs, voice data, and social networks to try and lower toxicity in esports and improve the experience for the individual player. We try to understand how toxicity emerges and propagates in the community and what prompts players to become toxic.

Teamwork research in esports has been around for a while, recognizing that esports teams are operating in an incredibly stressful environment with high-performance requirements. Just like any other sports team! From team matching to team operations, esports analytics combined with behavioural psychology have become a source of information about how teams are formed and function, which is set to have a substantial impact outside of esports.

On the topic of game design, the research has focused on how to design esports and how to maintain them. Maintaining control of the evolving adjustments needed to keep these games operational requires intense live operations, as recently documented by Sunny Thaicharoen who presented the first model of the metagame of esports, This model shows how knowledge flows among esports stakeholders in fascinating action-reaction loops, as players react to game updates and develop new strategies. Esports comprise a sizeable fraction of the 220bn dollar global games industry and one of the fastest-growing components. This combined with the data-driven nature of esports means that data-informed design is of perennial interest.

Then we have audience experiences and esports production. This is a relatively new endeavour on the academic side of things, championed by the Weavr project from 2017 onwards. Up until that point, esports analytics research largely ignored audiences, but with the enormous Weavr project, that saw seven esports companies joining forces with academic experts, an incredible push was made for data-driven tools that would on one hand help broadcasting teams and online creators, from live-streaming to studio-based coverage, and on the other hand tools and models that would enable esports content to be enhanced with data.

For example, Weavr developed applications that provided a personalised, data-driven feed of live matches for the individual viewer, supported by an array of machine learning models all operating in real-time – across screens, AR and VR.

Finally, we have education. Esports engage a lot of young people, and just playing these games carries lessons about teamwork, communication, strategy and more besides. This is being harnessed by schools, youth clubs, camps, therapists, and more worldwide. Central to such efforts is again data and analytics, and the intrinsic motivation to engage with data, and learn, created by these games, which is perhaps the most central, positive societal impact that esports has had.

I am old. Too old, according to my children. I remember the first academic conferences we had on technical research in games, and how much esports like Starcraft played a role in kickstarting technical games research more broadly, esports analytics specifically, and our whole community. Esports remain very visible in the academic games research environment today, with AI, data and analytics being at the heart of many endeavours, playing a supporting role in a variety of academic fields.  What makes me very happy to see is how our work is now moving beyond esports, beyond games. There can be no finer purpose for scientists than trying to create a positive impact on our world. You might even question if any other purpose can be justified.

If you would like to know more about esports research, the Esports Research Network is an international and welcoming environment for academics and others working with esports research.  The Arena Research Cluster is also an international research network, but focused on technical esports research.

If you would like to know more about how games are shaping society and how we can ensure that games as a phenomenon lead to societal benefit, check out the white papers on the Digital Observatory Research Cluster.