Understanding Player Experience Through the Use of Similarity Matrix

The paper proposes a similarity matrix method for analyzing player performance in games, adapting techniques from computer science fields such as text, video, and music analysis to the study of gameplay. The approach converts gameplay audio into chroma representations, discards the video channel, and computes pairwise similarity between sound segments. Diagonal patterns in the resulting matrix reveal sustained sequences of similar activity, making it possible to visualize the structure of a play session. The method is positioned as an extension of earlier feedback-based gameplay metrics, with the aim of offering game studies a new analytical lens on how players experience and move through games.

The central findings come from applying the technique to recordings of three titles: Max Payne 3, Battlefield 3, and Super Hexagon. The authors report that similarity matrices distinguish player strategies, such as players who spend time customizing settings before playing versus those who immediately begin acting in the game. Repeated diagonal patterns identify replayed sequences, including those following player deaths. Skill differences also appear in the visualizations, with more proficient players producing long, unbroken stretches of similarity while less experienced players generate numerous short diagonal segments. The paper concludes that the method is a valuable addition to game studies and points toward future integration with controller input data and biometric measures.

The scope of the work is focused on a small set of commercial titles and recorded play sessions, spanning action shooters and a rhythm-driven arcade game, with analysis centered on audio rather than visual or physiological signals. The paper's three planned applications, comparing players within a single game, detecting repetition within one performance, and tracking progression through soundtrack similarity, are described as directions for development, and the early sections present the method's potential more than quantitative validation. Taken together, the work establishes a promising, exploratory framework for player experience research that is grounded in established research on gameplay metrics and player modeling.