The biggest Steam Labs project, however, is the Interactive Recommender. Valve says that Steam’s greatest asset is its massive catalog of games. But having so many titles makes it hard for customers to sort through it all and find the games relevant to their interests. That’s why Valve is introducing a new, machine learning-powered method to get the right games in front of potential customers.
How the Interactive Recommender Could Fix Steam’s Search Problem
Currently, Steam relies on a tag-based system that recommends games to customers based on titles with similar, user added tags. But the new Interactive Recommender will suggest games to customers based on a user’s playtime history and “other salient data.”
Valve says the new system doesn’t use tags or genre data at all, and only explicitly gathers the release date of a game. The Recommender learns the rest of the information it needs during the training process. Valve says that not including genre or tags yields better results, though we’ll have to test Steam’s Interactive Recommender ourselves to judge.
Players will have access to tweak their Interactive Recommender, choosing to let the Recommender focus more on newer or older games; and on either mainstream titles or more obscure ones. Valve says its experimental Recommender tool will be able to help all customers, whether they want the newest game releases, or hidden gems.
Valve also addressed how because the new Recommender relies on machine learning, the model can’t recommend brand new games that customers haven’t played yet. Although the Recommender can learn quickly, Valve says it’s keeping the existing Discovery Queue system to help surface brand new games. The Recommender will serve as an, “additive to existing mechanisms rather than a replacement for them.”
Indie Devs Say Getting Good Recommendations on Steam is Broken
The Interactive Recommender appears to be geared towards solving Steam’s search problems. Especially with how Steam seems to be prioritizing bigger, triple-A games at the expense of indie games.
In October 2018, Valve made a change to Steam’s search algorithm that gave more weight to sales and wishlist activity when recommending games to customers. This led to a situation where recommendations, in Valve’s own words, “had the unintended side-effect of de-boosting tags in the ‘More Like This’ section on a game’s store page.” The result was that many indie game developers saw a drastic decrease in traffic to their games’ store page.
Developer Jake Birkett was one of the first to publicize the damage done to their game sales as a result of Steam’s algorithm changes. “In the past I have felt positive about Steam, but these discovery changes and the recent revenue share changes that are only relevant to hugely successfully games don’t make me feel particularly positive about the future of selling games on Steam,” Birkett wrote on his personal blog. “In fact I’d go as far as to say I’m worried.”
Although Valve says it made changes to the recommendation system that should’ve fixed the problems from the October bug, indie devs say game sales never fully recovered. Even as recently as the past Steam Summer Sale, developers say that traffic from Steam’s internal discovery queue have been weaker, ever since the algorithm changes made in October 2018.
Indie developer Yitz told Kotaku that trust in Steam has weakened since October 2018. “This Steam sale was a disaster, but I’m far more concerned about the overall trend we’ve seen in the Steam algorithm since October last year: pushing unpopular (including ‘mostly negative’ reviewed) triple-A games over titles that Steam has more than enough data to know would be a better match for the consumer.”
Steam’s new Interactive Recommender has just launched, and it’s only available to users who opt-in to test out the experimental new feature. But we’ll be keeping an eye on how it potentially changes the way games, particularly indies, are discovered on Steam.
Matt Kim is a reporter at IGN. You can reach him on Twitter.