Steam’s experimental Interactive Recommender tool uses machine learning to more accurately glean the games Steam users might be interested in, and Valve says the early days of the Steam Labs project are showing promise.
So far, Valve says that nearly 10,000 new games have been wishlisted thanks to Steam users poking around in the Interactive Recommender, though it notes that those results might be skewed due to the amount of publicity the tool has received since its launch last month. But, overall, Valve says the Recommender has so far performed very well in terms of measurable positive action like wishlists or purchases.
As outlined in the full blog post (alongside details about some of the other Steam Labs experiments), Valve is still tweaking and changing the Interactive Recommender to see what works best for everyone. UI changes based on user feedback have been made, along with a new feature that lets players exclude some of their recently played games from the tool’s consideration. And, with machine learning at it is core, the recommender itself is always adapting.
“The interactive recommender model adapts in two ways. First, it adapts right away to an individual user’s behavior; as you play new games, or revisit old ones, the model uses that data to give you updated recommendations. The second way the model adapts is by periodically re-training itself to take into account global changes, staying up-to-date with the latest releases and the gradually changing patterns of player behavior,” explains the post.
“This re-training process is an intensive operation that crunches billions of data points and can take a whole day to complete. We’ve been doing some behind-the-scenes cleanup work to make the re-training process smoother and more automated, which will enable us to use the technology in new contexts, like other Labs experiments or the Store itself.”