Valve has revealed a brand new feature for Steam called the Interactive Recommender, and while this sounds relatively standard, it becomes more interesting once you learn that it's using machine learning via a neural network that's trained to recommend games based on data like playtime history.
This feature was spawned from fan feedback that they'd like better tools to help find games, hence why machine learning is being brought in for "personalised recommendations", combining with the user's ability to adjust the results.
The neural network was trained using billions of play sessions and millions of users, with parameters like popularity and release time able to be adjusted. The model learns from the games itself, and release date is "part of the model training process, [which] yields better quality results than simply applying it as filter on the output".
Tags and review scores aren't used as information, so they can't affect results, but the model instead learns about games based on what the users do. That said, users can still filter by tag, but the model disregards most data like genre or price, looking instead at what games you play and what others play, basing its decision on others playing games. Players with similar habits to you will then inform your recommendations.
This is an experiment that people can use if they want, but it's not being the standardised change just yet. Valve sees this as "one discovery element among many, and look forward to introducing more ways to connect customers with interesting content and developers". The Discovery Queue, for example, will be an extra tool to help surface brand new content, since this Interactive Recommender won't have data on brand new stuff.
"We want to hear feedback from both customers and developers, so, check out the The Interactive Recommender and join the discussions to let us know what you think. As we gather data about the recommender's usefulness, we'll share how things are going," Valve notes, adding that developers can also see how many page visits are coming as a result of this new tool.
Does this sound like a good idea?