This is all part of Jennifer Dewalt's "180 web sites in 180 days" experiment: http://jenniferdewalt.com
I'm not trying to nitpick, this is an inspired UI. It's just that a spike in something like "joyful" or "scared" could be interesting to see in relative terms, especially when calculating the positivity index at the top. I know that statistical accuracy isn't necessarily the point here, but working with twitter data can be tricky. I'm sure that engineering it was hard enough. What an interesting visualization.
[1] For example when looking at a standard corpora like the NLTK movie reviews data set, the term "love" appears 1,588 times vs "helpless" appearing 47 times.
I do not know if normalization would help in this instance as people are taking a deliberate action in hash-tagging with the related feeling.
I've done some sentiment analysis and relationship/influence mapping on twitter data and the 140 character limit often means you have to create a specialized training set for your classifiers based on the group of people you are targeting. Simply weighting existing training sets offer very small benefits in accuracy.
Still, love the simplicity of this UI and concept.
Another common issue is context ... ie detecting "I feel awful ... Need a coke" as negative sentiment against coke.
This is far more rudimentary.
I feel like righ now I would want to relax and do nothing, but I know tat if I had the oppertunity I would get bored, but from the boredness I would start doing something interesting and creative.
Bored: "feeling weary and impatient because one is unoccupied or lacks interest in one's current activity."