Training a model for in game level selection based on player satisfaction

Abstract

Player satisfaction is among the greatest concerns for game developers. In this paper a system is proposed with the ability to select new levels based on the players satisfaction in former levels. Previous research has been done into this fi eld, often utilizing complex sensors or requiring additional player effort. This research aims to capture satisfaction with only in game measurements and no additional player effort. The focus is laid on developing a model that is not obstructive to the player and is easy to implement and understand by a game developer. This has been accomplished by utilizing multiple linear regressions and using user engagement as a measurement. The research has been conducted on the game of Sudoku and utilizes artifi cial players. The adaptation to artifi cial players comes a result of the COVID-19 pandemic, which made it impossible to let human participants partake.