About this project
Every game on Steam and Metacritic, placed in 3D space based on the tags real players apply to them. Games with similar tag profiles cluster together. Games that resist easy classification end up somewhere in the space between.
by Cooper Colglazier · games user researcher · spacebetweengames@gmail.com
How to use
Methodology
Data sources
Game metadata and review scores come from two primary sources: Metacritic (~14,000 games with critic and user scores) and Steam (tags, user reviews, and store data via SteamSpy and the Steam Store API). The combined dataset covers approximately 80,000 games.
Tag vectors
Each game is described by a vector of Steam user tags with proportional weights (reflecting how often the community applied each tag).
For example, Sleeping Dogs and Sifu are both martial arts action games set in crime-filled cities — and they end up near each other on the map. What they share:
But what separates them is what pulls them toward different neighborhoods. Sleeping Dogs has:
While Sifu has:
Numbers show how many Steam users applied each tag to that game (Sleeping Dogs / Sifu for shared tags).
Tags like Base-Building or Detective carry more signal than ubiquitous ones like Singleplayer. Around 72,000 games have real Steam user tags. For the ~9,000 Metacritic only titles without Steam presence, tags were predicted using an AI model. Games with fewer than 4 tags lack enough information to be placed meaningfully and were excluded from the map. Roughly 1,500 utilities and audio/video production tools from Steam were also excluded.
Dimensionality reduction
The high-dimensional tag vectors are projected into 3D space using UMAP (Uniform Manifold Approximation and Projection). This preserves local neighborhood structure: games that share tag DNA end up near each other. Developer and release year features provide a secondary clustering signal, so games by the same studio or from the same era tend to group loosely together.
Tuning UMAP parameters is part science, part art — balancing readable spread between games against tight, accurate clusters. Understanding UMAP is a useful resource for learning how parameter choices shape the final layout.
Don't @ me...
- / This project is not affiliated with Metacritic, Steam, or any game publisher.
- / Review scores and metadata may not be perfectly up to date or entirely accurate. Data was collected at a point in time and may drift from current listings.
- / AI-predicted tags (for Metacritic only games without Steam presence) may definitely contain inaccuracies.
Spatial placement is purely algorithmic (and stochastic — running the data pipeline again would produce a slightly different mapping). In other words, the specific clustering and placement of games shouldn't be considered perfect. Games that are near each other share similar tag profiles, not necessarily similar quality or feel.
Genre-blending games naturally end up between clusters. For example, Grand Theft Auto IV sits almost exactly equidistant between the RPG and Shooter neighborhoods — it shares DNA with both, so the algorithm places it in the space between.
Distance on the map is not a quality judgment. Importantly, distance between clusters and specific games is mostly meaningless and an artifact of the algorithm. The intention with this project is to help people visualize the gaming space and find new games via natural exploration.
Inspiration
This project is inspired by Every Noise at Once by Glenn McDonald — a genre map of the music world.