Steam Labs is a virtual place where Valve conducts "experiments" aimed at simplifying the use of Steam. One of the most interesting experiments that were still an interactive adviser, he uses machine learning instead of ordinary tags to find games that may interest you.
The latest experiment "What to play?" It performs a similar function and is actually based on the same technology as an interactive adviser. But instead of looking at the store to find games that you may want to buy, it is based on games that you have not played yet.
on the page "What to play?" Three games from your library are listed, which you may think about. Each with microtreiler (another experiment in Steam Labs), tags and a short list of similar games to give you an idea of this game. If you decide that you know better what you want, that is, the "next choice" button. It is not clear whether there is a strict restriction for recommendations, or whether there will be the number of proposals to vary depending on the size of your library.
Valve warned that there may be some inaccuracies when using the experiment, since they continue to configure the system. "The algorithm that we use to select comparable games is also a new system with which we experiment in this laboratory, and this is an incomplete process. Therefore, do not be surprised if he sometimes shows some unexpected points of comparison, "writes Valve.