+1 for #3. I was working very actively on that problem about a year ago. The issue is to scale such thing with current way. Also typical chicken and egg problem
I think the scale issue is a technology problem... it's finding the right algorithm or model that scales itself vs. being human driven. You might be able to do the latter for a while to build the audience, though, and slowly replace it with the former over time (to help with chicken / egg problem).