Engagement-Based Game Level Adaption

Making game levels optimally engaging for every player is a perennial challenge of game design. Most work in this area revolves around balancing the difficulty to match the players skill, either through laborious pre-release testing and tweaking or dynamic difficulty adjustment during play. However, difficulty is just one design aspect that affects engagement, and there are many games types where standard difficulty balancing techniques don't work. A good example are games with millions of user-generated levels such as Mario Maker or Little Big Planet. Here, the features of a given level are unknown in advance, and a player has to search through many levels to find one which best suits their skill level and preferences.

To solve this problem this project brings together game design and artificial intelligence (AI) to develop a system that automatically estimates a player's skill level and preferences and which level from the large pool will best fit this profile, creating a personalised recommender system. Longer-term, this system would also be able to provide companies with greater insight into their player’s preferences and the likely engagement performance of levels they have in development without having to actually playtest them.

Our aim is to develop AI middleware, built on validated psychological theories of engagement which can be used by companies to optimally engage users and support creators.