It's harder than it seems. Flagging also opens the door to 'pranksters' that will false-flag everybody just to clog the system. Most of these things seem to be 'obvious' until you actually start to implement them in real life and then it seems that old saw about 'theory and practice' holds very true.

Flagging also opens the door to 'pranksters' that will false-flag everybody just to clog the system. Seems to me this could be mostly avoided by having the flagging mechanism repsond to some # > 1 of flags from unique users. Issues could still come up if there were > 1 people engaged in this practice simultaneously, though it doesn't seem like an activity holding much long term appeal. So, this solution seems while perhaps not optimal at least functional.

Please share some other hard type problems from such sites if you feel like it. Then we can talk through/theorize about what might work in practice.

Also, is someone keeps flagging people, then there flags don't count. This has been solved before.

However, someone who uses chatroulette may have good reasons to flag every person they see as a pervert.

That's exactly the key, abuse patterns and real use patterns are so similar that it is very hard to automate the analysis.

> Please share some other hard type problems from such sites if you feel like it.

I'll work up a list of stuff, over the years I've spent many hours analyzing the abuse angle of webcam sites and while I haven't found a completely automated way to do it we have it worked out to the point where with relatively simple tools a single person can easily monitor a few thousand cams. Chatroulette is one-to-one, which has a bunch of unique problems in terms of the speed with which you can gather statistics on cams in order to determine the kind of content.

Ironically the best indicator for mis-labeled adult content is the number of viewers...

If they did that surely there'd be hardly anyone left on the site.