It's called Steam Labs, and the first bits of mad video game science to emerge from it are promising.
Valve says the new system doesn't use tags or genre data at all, and only explicitly gathers the release date of a game. Valve's recommendations have historically only worked on matching the genre and tags of games you've most recently played, so to diversify the field it's pulling from should hopefully allow players to find a wider variety of games as well as give the spotlight to some indie devs who aren't usually included in recommendation lists.
Now in a rare piece of (potential) good news for indies, Valve has today announced three new Steam Labs features, all created to help players discover a wider variety of new games. It is a machine-learning tool that can recommend games based on your playing habits - that alone wouldn't be worth writing home about, but the feature stands out from traditional algorithms by letting you customize it.
The algorithm uses the information to suggest new games to the user based on the extracted information. For example: it can be directed to popular titles released in the past ten years, or niche games launched in the past six months. Clicking on a collection will bring up a grid of videogames, you can then hover the mouse over one title to start the trailer or over the entire row to start them all. Steam, which is now the largest PC gaming client, gives users recommendations on what to play next. It is essentially a six-second trailer for every Steam game.
The final experiment Valve is testing out is called "Automatic Show", which the company jokingly compares to "one of those cable shopping channels without the super-absorbent chamois cloths".
The micro trailers page asks the question "What if you could absorb every game on Steam in just moments?" and shows off a range of trailers in various categories, like roguelikes, visual novels, and "dragons".
The second, the curiously named Interactive Recommender, is the one that's likely to be of most interest to indie developers and publishers. Valve boasts that the model is based on millions of Steam users, and you can filter results with a series of sliders. The automatic show is... Valve also promises the Recommender won't force developers to game the system with specific tags, pricing, or advertising styles, although without knowing more about how it works, it's hard to rule out users discovering quirks in the engine. We sometimes see as many as 30 new releases for headsets in a week.