In 2005, Jonathan Harris and Sep Kamvar created We Feel Fine, a project that searched blogs for mentions of
feelings and visualized them in a Java applet written in Processing.
Their complete methodology is still available online.
I thought it would be fun to update the project for
2013 by using the Twitter Streaming API, Node.js, and D3.js. I've followed We Feel Fine's original methodology pretty closely.
Check out the source on GitHub or just read this summary of my methodology:
- I'm listening to the Twitter Streaming API for tweets that contain
the feeling indicators: "feel," "feeling," or "felt." The inclusion
of the past tense is a divegence from We Feel Fine's methods.
- When a new Tweet comes in with a feeling indicator word, I create
a string of eligible words that starts two words before
the feeling indicator and ends five words after it. This is a
departure from We Feel Fine, which looked for eligible feelings
in the entire sentence in which the feeling indicator was detected.
Limiting eligible words to those closer to the
feeling indicator seems to provide higher-quality feelings.
- I search the string of eligible words for any word in the list of valid feelings that We Feel Fine compiled.
- If a valid feeling is found in the eligible words selection,
the source Tweet is sent to the browser, which draws a circle
representing the feeling tweet in the color that the original We Feel Fine
project used for that feeling. Sad feelings are blue or gray, happy feelings are
yellow or orange, loved feelings are pink, and angry feelings are red.
Who made this?