jemhoff

Good stuff! I wonder if you could somehow classify the conversation trees using depth or branching factor or whatever to predict and/or discover hot topics. Like you said, AMAs tend to have a pretty obvious structure but maybe that could be extended to discover "controversial" threads -- or even people!

How I'd do it: Use Embedly! Our extract API returns the dominant colors of all the images it finds: http://embed.ly/docs/extract/api/response#extract-images

For those that are curious, what happens under the covers is a random sampling of the image followed by a conversion to LAB color space (even better than YUV in terms of human perception) and finally some K-means clustering.

To see this in action, check out http://tightro.pe.

That feature is sweet. And tight.

Hey bro your red warning is blue, just sayin.

Whoops, you're right! I don't think I've ever noticed that before -- just changed it to a monotone cubic and it looks much better. Thanks!

http://en.wikipedia.org/wiki/Monotone_cubic_interpolation