Social networking data can provide lots of information, and fast. Plotting the prevalence of a keyword from among the population of tweets is an interesting way to normalize for the old problem of first order trends in a dataset. The raw tweet point locations of bored and excited would just look like a sampled distribution of tweets in general, by mapping proportion within equally populous zones you get a sense of actual variability. It's also interesting to choose a set of related keywords so you get a comparative graphic of related (in this case, opposite) cultural phenomena. In this case it turned out to be the relative enthusiasm of the US in a random day in the spring.
Each zone represents an area of roughly 10,000 geo-located tweets over the course of one day (May 11 or thereabouts, I think). This method was the idea of Eric Fischer, who mapped global subdivisions of equal tweet frequency. We set the number of tweets per zone to 10,000. Aside: in time, I'd like to play with the idea of an algorithm that stepped through increasingly higher tweet numbers to see if the stepwise process of large to small buckets might make more spatially compact zones (zones of super-dense content tend to be understandably narrow along one dimension but really long in the other). Daniel Briggs here at IDV used Eric's script to generate the boxes using one day of data from Twitter's sweet sweet API.
Using this underlying map as a denominator representing the general tweet population, we picked out tweets containing "bored" or "excited" and thematically mapped their proportion in Visual Fusion.
We're really interested in digging into the spatial power of social media to find structure in events as they happen. This shows the utility of aggregation, but here's another example where physical movement is illustrated with discrete tweets as seen in the context of geography and time.
Tweet Follow @JohnNelsonIDV