Steam’s “Steam Labs” program has a few interesting new experiments for players to try out and offer their feedback on. Perhaps the biggest is the “Interactive Recommender” system. This Recommender has a neural network model, trained to recommend games based on a user’s playtime history, along with other data. Based on data from millions of users, and billions of play sessions, this will ultimately give results that capture the various play patterns of their users, across the Steam catalog.
Steam does not feed the model information about the games, but instead, the model learns about the games for itself during the training process. They do not use information about tags or review scores either. Reviews or tags do not affect results, so players that are review bombing games will not affect your interactive recommender. However, just because you play a lot of Beat Saber does not mean it will only offer up VR rhythm game s- the model has a different approach. Instead it looks at what games you play, and what games other people play, and makes informed suggestions based on other people playing games on Steam.
Steam Labs can be found here, where you can see the current active experiments: Micro Trailers (six-second trailer for every Steam game), Interactive Recommender, and Automatic Show (automated, daily half-hour show about Steam games). They welcome feedback, and that can be left here.