Locating a person via Twitter is usually only possible if that person uses geotags in their Tweets. (Geotags in Twitter are basically tags that provide your geographic location.) However, the boys at IBM managed to bypass the geotag system and were able to locate where people live, simply based on their tweets — tweets that have no geotags.
This is possible because IBM assesses the similarities of the content in a user’s tweet with other users who use geotags within their tweets. According to Arxiv paper, this procedure makes for a decent location prediction tool.
You see, tweets can have several salient information that points to a user’s location. For example, Check-ins, Foursquare, Hashtags and text references relating to certain states or cities. By analyzing the content of tweets with geotags attached, then searching around for similarities in content tweets without geotags, researchers could come up with an idea of where a particular person originates from.
To train the algorithm, researchers used 90 percent of a body of 1.5 million tweets. The 10 percent is used to test the algorithm to make sure it works. Surprisingly, this thing has over 58 percent accuracy. Now, this is a far cry from being spot on, but with a little extra work, it is possible for researchers to pinpoint the exact location by working with the information they have.
Folks should not worry though, as location data in Twitter is usually used to feed ads, so don’t expect a naked gentleman in a trench-coat knocking on your door any time soon.