For years now, Valve has been testingnewapproaches to filter the glut of Steam games down to the ones in which individual users are most likely to show an interest. To that end, the company is today rolling out a machine-learning-powered “Interactive Recommender” trained on “billions of play sessions” from the Steam user base.
In the past, Steam has relied largely on crowd-sourced metadata like user-provided tags, user-curated lists, aggregate review scores, and sales data to drive its recommendation algorithms. But the new Interactive Recommender is different, Valve says, because it works without any initial internal or external information about the games themselves (save for the release date). “Instead, the model learns about the games for itself during the training process,” Valve says. “The model infers properties of games by learning what users do, not by looking at other extrinsic data.”
Your own playtime history is a core part of this neural-network-driven model. The number of hours you put into each game in your library is compared with that of millions of other Steam users so the neural network can make “informed suggestions” about the kinds of games you might like. “The idea is that if players with broadly similar play habits to you also tend to play another game you haven’t tried yet, then that game is likely to be a good recommendation for you,” Valve writes.
This, in turn, should pre-empt issues with developers trying to game the system by choosing popular tags or leaning on positive reviews, as they have with previous recommendation algorithms. “The best way for a developer to optimize for this model is to make a game that people enjoy playing,” Valve writes.
Individual users can tweak the parameters of these AI-powered suggestions to favor games released in a certain time period or games that fall to one side or another on a “popularity” gradient. “We’ve found that, especially for people who play a lot of games, digging into the ‘niche’ end of the range can be a very effective way to find hidden gems,” Valve writes.
Valve admits that this machine learning system isn’t ideal for brand new games, which don’t have enough players from which to glean data, leading to a “cold start” chicken-and-egg effect. Existing systems like Steam’s Discovery Queue should help get those titles in front of their first players, the company writes.
The new recommendation engine is joined by two other experimental products coming out of the newly launched “Steam Labs” brand today. A Micro Trailers page automatically generates six-second video vignettes to represent titles, arranged by genre, while The Automated Show will collect 30 minutes’ worth of footage from the latest Steam releases for easy consumption.