“Recommendation systems are self-fulfilling,” says James Allen-Robertson, who studies digital media at the University of Essex, UK. People who rely on them end up in a bubble, with the system only showing them games or films like ones they have already ...
Just so many stellar releases to choose fromFrederic J. Brown/AFP/Getty
The constellations spin dizzyingly, 15,000 stars against the blackness. Click, zoom in, and individual dots pop up from the nearest clusters: Everybody’s Gone to the Rapture (2015), Red Dead Redemption (2010), Nancy Drew Dossier: Lights, Camera, Curses (2008). This is GameSpace, an experimental online tool designed to help you find the next video game to play. It won’t just work for gamers, though – it could soon make life a bit better for anyone looking for the next great book or movie.
Like the rest of us, gamers can’t keep up with all the new titles constantly being published. “There are thousands and thousands of games,” says James Ryan at the University of California, Santa Cruz. “The accumulation is ridiculous.”
Apple’s App Store contains around 800,000 games, with several hundred new ones added every day. Even if they’re great, many will get lost in the crowd.
So Ryan and his colleagues developed GameSpace as a better alternative to the recommendation algorithms we get tips from: “People who watched Breaking Bad also watched Better Call Saul.” These often work well, with one big caveat. “Recommendation systems are self-fulfilling,” says James Allen-Robertson, who studies digital media at the University of Essex, UK. People who rely on them end up in a bubble, with the system only showing them games or films like ones they have already enjoyed.
“We wanted to build a tool that cuts through the morass,” says Ryan. “We wanted to subvert recommendation systems and capture how people actually talk about games.”
Instead of relying on the opinions of those who might also be trapped in their own filter bubbles, GameSpace drops you into a galaxy where every star represents a game, and similar titles are grouped into constellations.
To do that, the team used natural language processing software to scour 21,456 descriptions of games on Wikipedia. A machine learning system then identifies 200 points of comparison between descriptions to generate a similarity score measuring how alike any two games are. This score is what the team used to position the games in the 3D visualisation.
The tool has already thrown up some surprises. Why was a game about breakfast cereal sitting next to Doom, for example? “I thought the algorithm was broken,” Ryan says. But then he read the cereal game’s description and saw it was indeed an adaptation of Doom. “Not only was it not broken, but it was really cool the way it was bringing up related games,” he says.
One shortcoming is that the tool is limited by what people happen to have posted on Wikipedia. Japanese interactive novels are really well represented, says Ryan, but sports games are not. “They’re super popular, just not among Wikipedia authors,” says Ryan.
The team will present the work at the Foundations of Digital Games conference in Cape Cod, Massachussetts, next month. Extending it to books and films should be possible for any title with a Wikipedia description available online, says Ryan. You could even build a tool that let you fly through Wikipedia itself, he says.
More on these topics:
video games,online shopping,algorithm,Wikipedia,machine learning,technology,computing