New ‘Geography of Hate’ project maps hate speech on Twitter

A new project called ‘Geography of Hate‘ gives detailed information about the geographic distribution of hate speech on Twitter. According to The Verge:

the interactive map charts ten relatively common slurs across the continental US, either by general category or individually. Looking at the whole country, you’ll often see a mass of red or what the map’s creators call a “blue smog of hate.” Zooming in, however, patches appear over individual regions or cities; some may be predictable, while others are not. . .

Unlike many other studies, for example, the tweets weren’t collected and analyzed algorithmically — a method that could accidentally collect non-derogatory uses of these terms. Instead, the team first searched through a year’s worth of geotagged tweets for words, then had a group of students at Humboldt State University look at each one. Only tweets they found explicitly negative went on the map: a derogatory use of the word “dyke” would be added, for example, but one reclaiming the term for a gay pride parade would not. In total, the map charts about 150,000 negative, slur-filled tweets.

Since the map looks at only geotagged tweets, it’s not a pure representation of Twitter, but this is standard practice for such mapping. Hateful tweets are weighted by the total number of tweets in an area, so you’ll see the proportional number of slurs, not just areas with the largest number of Twitter users.


The information is incredibly interesting (and eye-opening!), the map is user-friendly, and there’s loads of information available about the study’s methodology. Go check it out!

8 thoughts on “New ‘Geography of Hate’ project maps hate speech on Twitter

  1. It’s unfortunate that they didn’t control for population density. There are a lot less hateful tweets coming out of Montana than New York, I suppose in part since there are a lot less people in Montana than in New York.

  2. Sch, they didn’t look at population density, but they did look at ‘Twitter density’ (if that’s a thing. . .):

    “Hateful tweets are weighted by the total number of tweets in an area, so you’ll see the proportional number of slurs, not just areas with the largest number of Twitter users.”

  3. Anyone analyze the data yet? After looking up my own area and seeing that the suburbs have tweets, some of them slightly-light-blue on various categories, I wonder if there is any more interesting conclusion that could be drawn from a more careful look.

    I noticed that at my suburbs/university student areas, terms like gay and fag are rare but dyke is more common. I wonder if it was being used as hate speech (but not the other terms?) or if it is just an artifact of the local LGBT twitter community, which would use the term but not as hate speech. Oh well, back to work.

  4. The tweets deemed hateful by the team may have been weighted by total number of tweets in an area, but for whatever reason the map does bear more than a passing resemblance to color coded population density maps of the United States as well as to density maps of Twitter usage.

    There are some unusual features to the map. For example, as near as I can make out, three of the biggest hot spots in the Midwestern United States appear to be the alentours of Sigel, IL (pop. 386), Kalona, IA (pop. 2,363), and tiny Muscatine Municipal Airport, IA. I wonder if a very small number of Twitter users with a high proportion of tweets deemed hateful, at least in an area without heavy Twitter traffic, can skew this.

  5. T.M.: they also tried to correct for non-derogatory uses of the word.

    What I wonder is the extent to which use of popular slurs actually tracks hatred. I have met a number of people who I would call homophobic that would be completely devastated if their child called a gay person queer. And they seem to be even more shocked when people who are queer call each other queer! I am not an American, but I am told there is a similar trend with the use of the N word, where white people try and police the use of the word by black people. I imagine that there is an age/generational/liberal(?) kind of selection that is taking place here as a result (even if the medium wasn’t twitter).

    Also: not all uses of a slur should be considered equally hateful (or at all). I am particularly thinking of cases where a slur is used “explicitly negatively” between people who belong to the group it is usually intending to be harming. EG The N word can be used in a mildly negative way without intending or causing what I imagine the map is supposed to be tracking. I wonder if this causes the existence of “hate” words to increase in places where the population of people who use these sorts of pseudo-reclaimed words exists.

  6. Nemo, I think a lot of the resemblance to color-coded population density maps disappears once you zoom in a bit on the map. For example, there’s a spike in the pacific northwest, near Vancouver and Portland. But when you zoom in, you see that the there’s actually very little of the identified hate speech going on in those major cities (where most of the people in the area live), and the spike is concentrated further inland (where the population density is much lower).

Comments are closed.