GameSpace is a visualization of the videogame medium as an explorable 3D space. Each of the nearly 16,000 stars in its galaxy represents an actual game that exists in the real world, and stars are placed in the space such that more similar games are nearer to one another. For example, games from the same series might be positioned in star clusters, and clusters like these may themselves cluster together to compose larger nebulae corresponding to game genres or subgenres. The coordinates for the games in the visualization were determined by applying a combination of techniques from natural language processing and machine learning to a corpus of 16,000 Wikipedia articles about videogames (more technical details appear below).

Fly around to explore the space; select individual games by clicking on them. (If you're not sure what the controls are, click the controller icon at the top of the screen.) Once a game is selected, you can view its Wikipedia article or watch a YouTube gameplay video, all without leaving the space.

GameSpace is a project by the Expressive Intelligence Studio at the University of California, Santa Cruz. Its design and development was led by PhD students James Ryan and Eric Kaltman, with mentoring and support from their advisors Noah Wardrip-Fruin and Michael Mateas. Undergraduates Taylor Owen-Milner and Andrew Max Fisher carried out much of the development work. Our background music is "Slow Lights" by Lee Rosevere (license). Like its companion tools—GameNet, GameSage, and GameGlobs—this project is part of the larger Game Metadata and Citation Project (GAMECIP), a multi-year joint initiative between UC Santa Cruz and Stanford that is funded by the Institute of Museum and Library Services (grant LG-06-13-0205-13). GAMECIP seeks to develop resources and best practices for videogame discovery, preservation, citation, and metadata creation.

GameSpace is critically enabled by two techniques—latent semantic analysis (LSA) and multidimensional scaling—which we applied to a large corpus of descriptive text extracted from Wikipedia articles about games. LSA works from the subtle premise that things that are described similarly are likely to be, in fact, similar. By submitting a corpus of 16,000 Wikipedia articles about games to LSA, we're able to automatically compute numerical scores between games that specify how related they are. There's one problem, though: our LSA model is high-dimensional—you can think of it as an abstract space with 207 dimensions—which means that it's hard to visualize. (Our answer to the challenge of visualizing our native 207-dimensional model is GameSpace's companion tool, GameNet.) That's where multidimensional scaling comes in—this is a statistical technique that allows one to convert high-dimensional spaces into lower-dimensional approximations of them (typically two- or three-dimensional approximations, since those can be visualized for humans). Using a particular variant called t-distributed stochastic neighbor embedding, we built a 3D approximation of our 207-dimensional LSA model. Critically, this algorithm attempts to position games in the 3D space such that they are near the games that they were near to in the 207-dimensional space. So while it's not a perfect representation of the full model, it works to maintain important features of the full model's topology. GameSpace, then, is simply this 3D space rendered in a game engine (with a first-person camera and flying controls), along with some extra features.

Python, gensim, scikit-learn, Flask, JavaScript, three.js, jQuery, bootstrap, virtualjoystick.js.

Beginning in March 2017, GameSpace will scrape Wikipedia each night to collect any new videogame articles that have appeared (or any existing articles that have been expanded beyond our minimum threshold of 250 words), and it will process these articles to automatically add new games into the space. This means that the visualization will be living and growing and evolving as the collection of Wikipedia videogame articles grows and evolves. If you're wondering if a particular game is in the space, you'll want to check to see whether it has a proper Wikipedia article (i.e., one that exists, and is at least 250 words in length). If it doesn't, you can author one for it, and the game will then be added into the space the next day.

Due to the large number of games in the model, we're currently generating YouTube queries for gameplay videos and hoping that the results are relevant. Of course, some games may not even have gameplay videos on YouTube (yet!). If you encounter an offensive video, let us know and we'll remove the link.

Yes. Stay tuned.

You can find us on Twitter as @flygamespace, or you can drop us a line via email.

Here's a pack of images that you can freely use. Also feel free to take and use any screenshots of the application in action. If you can't quite get the image you need, let us know and we'll try to provide it for you.

A full paper is still forthcoming. In the meantime, you can cite:

• Our short paper that introduced an early prototype of GameSpace in 2015:
  
  Ryan, J. O., Kaltman, E., Fisher, A. M., Hong, T., Owen-Milner, T., Mateas, M., & Wardrip-Fruin, N. (2015). Large-scale interactive visualizations of nearly 12,000 digital games. Proc. Foundations of Digital Games. [pdf]


• A longer paper that articulates the approach of using natural language processing in the field of game studies:
  
  Ryan, J. O., Kaltman, E., Mateas, M., & Wardrip-Fruin, N. (2015). What we talk about when we talk about games: Bottom-up game studies using natural language processing. Proc. Foundations of Digital Games. [pdf]