[SITZ]

I agree that getting all this browsing data from users will rather decrease the signal-to-noise ratio early on. But, that simultaneously provides a way to explore much more content on web that remains simply unknown to most people because of a lack of discovery medium. From personal experience, Most people share a mere fraction of stuff they discover online. So, even though that might find some rare gem on web, they will share only items which they would be rather associated with in contrast variety of things they might be interested in.

Although, I haven't ironed out exact details, IIRC, Deep learning techniques have improved significantly to make sense out of data if fed a huge chunk of it. Basic Video and Image comprehension techniques are pretty advanced too. I'd like to believe that this system will evolve to serve the purpose too.