The gaming community currently uses an informal classification of games into genres such as first-person shooters, real-time strategy games, etc. While this classification is generally accepted, producing a more formal taxonomy of game types directly from data has several scholarly and commercial advantages. These include providing a basis for analysis of age- and gender-related data, statistically meaningful grouping in critical literature, improved game recommendations on retail websites, and better evaluation of a game's market potential before production.

Mapping the mental space of game genres is challenging, both because it involves subjective evaluations and because there are many axes on which games can vary. We collected pairwise similarity metrics of games from game players through an online survey to build a large similarity matrix that is the projection of a high-dimensional space representing the unknown and hypothetical true mental space of game genres. We then applied previous techniques in manifold learning and psychology to the problem of reconstructing the most significant dimensions into maps that can be meaningfully interpreted. We believe this is the first application of these techniques to games and one of the first to work with conceptual (instead of physical) data.

The resulting maps arrange related games into spontaneously arising clusters that sometimes contradict current marketing genres. We analyze several of these clusters and propose both interpretations for these ``true genres'' as well as axes that game players appear to use in discriminating between them. Our initial results indicate that game players tend to primarily distinguish games not by traditional genres but instead by aesthetic and mechanics, which is closely related to how developers construct games.