The way I can see something like this implemented is to take a ratings system (like what 4v4music.com has). You build a profile of each npubs activities, most zapped tracks, ratings, etc. Now take that data and extrapolate the tags for those tracks. And now you have a database of people/likes.
You then compare your npub with others who's profile most closely matches yours (highest wot scores). Any of their top tracks not listed in your activities list are probably good recommendations for new music.
