Steam Introduces New Experimental Algorithmic Recommendation System
- New recommendation system uses a neural network based on playtime history.
- Two main adjustable metrics based on popularity and release date.
- Designed to assist customers and developers.
Valve has rolled out a new “recommender” within the experimental Steam Labs. The Interactive Recommender allows users to find recommendations for games based on three metrics: A user’s playtime history and two handy, adjustable sliders. The sliders, which are based on popularity (Also referred to as “mainstream-ness”) and age, are adjustable by users to fine-tune the recommendations given. To make the system more specific, tags can be added to further change which games do or do not appear under recommendations.
Many redditors have come out in support of the new recommender, talking about how it actually provides the games they want rather than whatever is popular in similar genres to their playtime history. One user, thomar, mentioned that it was actually returning indie games, rather than “popular” titles like the Discovery Queue, which is currently the primary way Steam gives out recommendations.
In a blog post about the new experimental system, Steam mentions that developers are also being given backdoor analytics about traffic to their game pages. The “Traffic Breakdown” will tell developers how many users found their page and what ways they did, allowing developers to tweak things like tags in order to better represent their game to their target audience.
If you are interested in checking out the new system for yourself, check out the Interactive Recommender here.