Steam's recommendations will show less popular games

SteamSteamed is dedicated to everything related to the Valve PC gaming service.

Last October, the Steam algorithm briefly gave way, filling in the "More like" recommendations sections of users with popular games while negatively affecting the independent under-the-radar that needed to be exposed. At the time, Valve called it a "bug" that it had corrected, but small developers complain that Steam's recommendations continue to drop them ever since. Now, Valve says that the recommendation sections have been changed to not prioritize larger games and to focus more on custom selections for individual users.

Valve began his announcement admitting that the recommendations were problematic. "Previously, when customers searched for games by browsing the recommendation line at the bottom of the home page or in the" More like that "sections, they did not see as many different games as we would have liked," writes the company. "In addition, we received many reactions that" Recommended for you "felt too biased in favor of the most popular games and did not feel very personalized."

So Valve went off to hunt for insects.

"We detected bugs, such as the" Tag-Like "section of the recommendation bulletin, which indicated that the top-rated games (a category that does not change very often) generated too much of what the players saw," Valve has written. "We changed that. We also found that, in some cases, our time scale used to calculate popularity was too narrow, making the unpredictability of some games unpredictable. We have therefore extended the period of time used in these calculations. "

These changes should now be available to all Steam users. Valve tested them on a small group of users prior to launch, and found that patches required users to click more frequently on recommended games and view more individual game pages.

"In these changes," Recommended for you "has become less focused on popular games and has shown more relevant games for individual customers," explained Valve about testing the algorithm. "It turned out that clients in the experimental group were more likely to click on the games presented in the recommendations section, at a rate that was almost 15% higher than the control group. Increased customization means that this section introduces a greater variety of games and that customer impressions are better distributed among them. "

There was also a 75% increase in the number of unique games visited, and a 48% increase in the average number of visits per match compared to a control group with the previous version of Mountain-Dew-Gugging. algorithm, who was in love with popular games.

For years, Steam has been unable to offer smaller games to users who might be interested in it. This frustrated the independent developers, many of whom now see the potential success of the platform as a mistake, even if they generated a pretty good craze around the pre-version of their game. The recommendations were a source of particular concern to developers. Between this algorithmic change and new experimental tools, such as an artificial intelligence that analyzes your playing time and helps you determine the next purchase, it seems that Valve is finally making a concerted effort to minimize the problem.

Valve thinks that these first results announce a positive change for Steam.

"We are encouraged by these results and have generalized them to everyone," the company wrote. "We continue to make changes and lead experiments like this to enhance Steam's existing capabilities, while exploring totally different ways for customers to find games they like."