Twitter sentiment analysis tests

This post collects some visualisation tests being done at LUST on sentiment analysis.

A list of emotions (inspired by Plutchik‘s theory of emotion) and corresponding words is used as a base for the analysis.
Various sets of geolocalized tweets were analysed in search of those words in order to match them to those emotions.

Adopting this method, a number of visualisation tests were performed – some of them are more schematic, others more expressive. Some of these experiments might inspire  visualisation solutions for the UrbanSensing platform.

Heat map of emotions in the Randstad area, where green is good, yellow neutral and red bad (tweets that didn’t match any emotion are displayed in grey):

Pics 130

Pics 131
Heat map of emotions in a configurable applet: adjusting grid size

us-java1

us-java2

Highlighting locations with majority of positive tweets:

us-java4

And locations with majority of negative tweets:

us-java3

And finally a more “free” experiment: here emotions drop like paint from the originating locations on the map. Each color corresponds to one emotion (fear, anger, sadness, happiness, disgust, surprise, boredom).

Screen Shot 2013-05-14 at 6.11.31 PM