The vast amounts of user-generated content on the Web produces information overload more frequently than enlightenment. How do we get the big picture on what people are thinking?
Twitter Weather reduces large quantities of text into meaningful data, visualizing the prevailing mood by rendering a weather-report-style display. Positive and negative feelings about a topic are mapped to a "temperature" from 0 to 100. Supporting Twitter Weather is a user-trained Bayesian classifier that aggregates and ranks emotional content on a topic.
Extending the technology to a new application, I built a Comment Weather bookmarklet for news sites. When reading an article, click on the bookmarklet to get the emotional weather of the comments at a glance. The comments are scraped, rated for sentiment and aggregated, and the temperature presented as an overlay.
I designed and developed the idea. Stephanie Bian wrote the initial classifier. I built on that and wrote the backend in Python, multithreaded to avoid the bottleneck of the Twitter timeline. This service fed visualizations I made in HTML and Quartz Composer (the latter you can see as the animation above).