How your tweets can reveal where you live
Published 24/03/2014 | 23:14
Think you can hide your location from Twitter? Think again.
Researchers at IBM have developed an algorithm that can predict your home address using your last 200 tweets – even when geotagging is disabled.
People's tweets contain a wide range of information relating to their location, from 'check-in' data gathered from location-based social networks like Foursqure to explicit references to the city or town they are in.
Some also include more subtle indicators like mentions of their favourite sports teams, or restaurants they have visited.
IBM researchers Jalal Mahmud, Jeffrey Nichols and Clemens Drews have developed an algorithm that predicts Twitter users' time zone, state or geographic region first – based on their pattern of tweets throughout the day – and then narrows the location down to their home city.
The algorithm is also able to detect whether Twitter users are travelling, which helps to improve the location prediction accuracy.
When users identified as travelling are eliminated, location prediction accuracy improves to 68pc for cities, 70pc for states, 80pc for time-zones and 73pc for regions, according to the researchers.
"Experimental evidence suggests that our algorithm works well in practice and outperforms the best existing algorithms for predicting the home location of Twitter users," they said in their report.
The researchers believe the algorithm could be used to create location-based visualisations and applications on top of Twitter. For example, a journalist tracking an event on Twitter may want to know which tweets are coming from users who are at the event, and compare them to tweets coming from users who are likely to be far away.
Meanwhile, retailers and marketers may want to track trending opinions about their products and services and analyse differences across geographies.
The team said that their algorithm could be improved in the future by searching tweets for mentions of local landmarks that can be pinpointed more accurately.