Augmenting Hearthstone Broadcasts with Real-Time Prediction and Move Explanations - project 2

Project Description

Hearthstone – a virtual Magic-like card game – has become one of the most popular eSports, with one of the highest ratios between watching and actually playing the game (predominant time is spent watching).

The relatively slow pace of gameplay and long thinking time between turns puts high pressure on commentators to make coverage interesting and engaging. However, in order to fully enjoy watching Hearthstone, one requires a significant amount of knowledge about a large pool of cards, deck building and general strategy. Particularly for novice viewers, this makes Hearthstone particularly hard to watch.

This project explores how, using machine learning and visualisation, conventional broadcasts can be augmented to appeal to broader audiences. In poker, small overlays showing win percentages gives even lay spectators an opportunity to anticipate what is most likely to happen and the be surprised if a player wins against the odds.

This principle could also help to make Hearthstone more enjoyable: if viewers can be helped to understand who is leading and what cards will most likely be played and why, a deviation and subsequent surprised can generate entertainment value. In order to achieve this, students will implement a simulation engine that understands the game rules and cards, and can be used to quickly simulate thousands of games and outcomes.

This knowledge will then be used to establish predictive models and visual overlays showing their outputs, such as win predictions, card draw probabilities, and a probability for each player to win on the next turn.

Number of places available


Required skills

(Both students)

Required: Good ability to develop in C# / Java

Required: Knowledge in Hearthstone (or significant in other deck-building card games, such as Magic)

Ideally: Experience with basic machine learning algorithms

Ideally: Experience in graphics design


Dr Florian Block

Dr Sam Devlin

Dr Nick Sephton


NJ Sephton, PI Cowling, SM Devlin, VJ Hodge, NH Slaven/ Using Association Rule Mining to Predict Opponent Deck Content in Android: Netrunner. In IEEE Computational Intelligence and Games Conference (CIG 2016)

S Devlin, A Anspoka, N Sephton, PI Cowling, J Rollason. Combining Gameplay Data With Monte Carlo Tree Search To Emulate Human Play. In Twelfth Artificial Intelligence and Interactive Digital Entertainment Conference (2016)


How to Apply

For more details on the summer school application process (including eligibility and funding) please see the overview page at: