Steam Leverages Machine Learning in New Interactive Recommender
This comes to us from your upcoming News Editor Poorna Shankar:
Steam introduced a new “interactive recommender” today, dubbed, um, Interactive Recommender. According to Valve, they’ve apparently heard our cries for better tools to help us discover games specifically tailored for the individual player.
Per Valve, “Underlying this new recommender is a neural-network model that is trained to recommend games based on a user’s playtime history, along with other salient data. We train the model based on data from many millions of Steam users and many billions of play sessions, giving us robust results that capture the nuances of different play patterns and covers our catalog.”
In essence, Valve is leveraging machine learning to recommend games to you personally based on data from millions of other Steam users. They then expose controls to you to determine how popular or niche your taste is, and how old or new the prospective title should be.
Valve claims that regardless of the combinations from the sliders, the end result should be game recommendations that are personalized and relevant to you. For example, by leaving the sliders at default my results are topped with Assassin’s Creed Origins and Far Cry 5 as observed in the image at the top of this post. However, if I weigh the popularity slider more towards niche, my results now include RAGE 2, the Dishonored: Death of the Outsider DLC, and Metro Exodus. Personally, these are all games I would still consider playing.
Curiously, Valve does not feed the model with data specific to games, stating, “Instead, the model learns about the games for itself during the training process. In fact, the only information about a game that gets explicitly fed into the process is the release date, enabling us to do time-windowing for the release-date slider. It turns out that using release date as part of the model training process yields better quality results than simply applying it as filter on the output.”
Valve doesn’t appear to be feeding tags or review scores into their model either, relying more on user behavior. Their reasoning is that they do not wish for review scores to affect the outcome of the results of the recommender.
Valve also appears to appeal to developers by noting that the model does not require developer optimization, “The best way for a developer to optimize for this model is to make a game that people enjoy playing. While it’s important to supply users with useful information about your game on its store page, you shouldn’t agonize about whether tags or other metadata will affect how a recommendations model sees your game.”
They conclude by saying that while this recommender is intended for consumers, they also wish to aid developers, “Developers can see how many page visits the recommender is generating directly in the existing ‘Traffic Breakdown’ page for each game, though note that this experiment might not generate much traffic relative to the rest of Steam.”
What do you think of Valve’s new Interactive Recommender? With Nvidia’s Turing GPUs featuring Tensor Cores, itself hardware for the DLSS machine learning feature to save on performance through AI-powered image reconstruction, what do you think of the increased use of machine learning in games?