Normalized Topic Heatmaps

Select a topic to make a heatmap from tweets matching that topic.

Topics are inferred using Latent Dirichlet Allocation. We labelled a subset of the topics with names to allow easy searching.

Use the mouse to pan and zoom the map.

Max number to plot controls the number of tweets sampled for plotting. Larger values will provide more accurate heatmaps, but increase plotting time.

Heatmaps in this view are normalized, showing how many tweets match the query compared to the total number of tweets from the same geographic area. The normalization is performed by dividing the Earth into into small rectangular sections and calculating the total number of tweets in each section. The weight of each tweet is then set to be a ratio to the total number of tweets in its section.

Only tweets that have an associated GPS location (about 10% of collected tweets, see the full paper for details) are plotted.