I think the real problem is Twitter refuses to make sophisticated filtering available to users. Simple mute, block, and keyword filtering are inadequate.
"Share because it makes me angry" etc
Seems like they'd get amazing whole text training sets, as well as a signal to better guide their platforms (e.g. under-weighting negative emotions in share velocity).
If a lot of people close to you in your social network react "angry" to something and you choose to share it, there's a good chance you're sharing it because it makes you angry too.
Graphs are cool
Makes me want to mess around with a browser extension that uses perceptual hashing to filter smug frog images or whatever.