How Did They Assemble That Google Map Of Hateful Tweets?

On May 10, 2013, Dr. Monica Stephens of Humboldt State University released a Google Map on the HSU website of geocoded hate tweets by region between 2012 and 2013. The map claims to demonstrate the variability of the frequency of hateful tweets relative to all tweets over space by aggregating negative homophobic hate words to a county level and normalizes it by the total Twitter traffic in each county. In laymen's terms, the map is a composite of racial, homophobic, and sexist tweets by state.

The purpose of the map is to explore social media as a conduit for hate speech and how intensely tied the virtual space is to socio-spatial contexts offline. Stephens received funding for the map from Humboldt University's Research and Creative Activities Fellowship. The map was constructed using Google Maps API software by Humboldt State undergraduate students Amelia Egle, Miles Ross, and Matthew Eiben as part of an advanced cartography course by Dr. Stephens. Additional information on methodology, research methods, and background can be found on Floating Sheep.

The map was constructed by the students who read and coded over 150,000 geotagged tweets that referenced hate slurs. By using this methodology, Stephens claims the map accounts for the manner by which tweets were intended, thereby avoiding algorithmic sentiment analysis or natural language processing as students were able manually discern tweet idioms:

"For example the phrase 'dyke,' while often negative when referring to an individual person, was also used in positive ways (e.g. 'dykes on bikes #SFPride'). The students were able to discern which were negative, neutral, or positive. Only those tweets used in an explicitly negative way are included in the map."

In terms of the study's shortcomings, Stephens admits that the map does not show any particularly revealing spatial distributions. In a related study by Stephens, which maps racist tweets by location following President Obama’s re-election, she also admits that as geocoded tweets are products of GPS-enabled smartphones and are more likely to originate from wealthier locations. The map may therefore under-represent rural and poorer locations.

However, the map seems to be an accurate indicator of social-geographic sentiment. The map aligns when homophobic tweets are cross-referenced and then compared to a cartogram that illustrates the legal status of same-sex marriage and civil rights by region in the United States.