Looks very cool. It seems to imply that, while you are modeling correlation between participants, you're ultimately going to wind up with a correlation distribution, and an expected correlation, and therefore, a guess.
Who knows which way sentiment will migrate as participants begin to copy one another. I do find the idea of modeling such things fascinating, though.