This might run into conflict with nuanced tastes, e.g. I would be a miserable gamer if this is how games are recommended to me because I actually really don't like very shiny new triple a shit. Spiritfarer changed my life and made me weep ugly tears multiple times in a way God of War never did, but I don't think most people would enjoy the gentle and tender approach it has towards its subject matter. [In Spiritfarer, you are a boat captain who picks up and hosts dead spirits until they ask you to take them to the gate where they are gone forever; in the meantime they share their lives with you and mull over death with you. Your only task is to care for these people until they are ready to go.]
>>Eating+(OP)
I think this would help people find more games like Spiritfarer if you can cut through the obligatory noise generated for the big AAA titles.
Imagine if you had the dataset to say "remove every reviewer who ranked God of War above Spiritfarer" you would probably be left with an amazing set of recommendations.
>>legits+p1
At that point, is it different from a general recommendation engine of users who like what you liked also liked X? [A thing which I also struggle with because I can never search based off of specific qualities no matter how many tags I search and exclude in Steam...]
>>Eating+i2
Yes. It would be an order of magnitude more accurate. Current recommendation engines currently have account for lots and lots of false positives and are generally constrained to shared product tags.