Geotagging One Hundred Million Twitter Accounts with Total Variation Minimization

Although many users choose not share their location information, there are methods to estimate the city of a Twitter user with a high degree of accuracy. Social ties in online networks often reflect geographic closeness in the physical world. So even if the individual has chosen not to share their locality, the location of their common contacts can reveal their whereabouts. This is the principle behind a new method for inferring the city of Twitter users.

Compton et al. have designed a method that estimates the location of a large number of related users. This approach relies on multiple sources of public information, depending on how much detail each user has revealed:

• Users that have activated the GPS functions on their devices provides location information directly.
• Users that have reported a ‘home’ location in their Twitter profile provide approximate location information.
• Users that have declined to share location information, an algorithm infers their location based on the location of their friends who have made their information public. Friends are determined by the number of reciprocal mentions between users – that is, how many times two users tag one another in tweets.

The method isn`t perfect, providing poor information on users who move often and far, or on those with a geographically dispersed network. However, the method can provide accurate location information by factoring in individual movement and the spread of the network. From over one billion mentions between 110 million users, the algorithm correctly identifies the tweeter’s city in 89.7% of cases. This is an improvement on other known methods for deriving location information.

Mining publicly available data in this way can leverage the connections between a large volume of social media contributors to support better understandings of activity and behavior in physical space. Even a user who takes steps to keep private their whereabouts can be found through the location information of their network. This research provides a potential method of deriving the approximate location or movements of twitter users based on their contacts. This approach can assist with attribution by increasing the effort required to conceal a social media user`s identity.

Vous pourriez également aimer

Mots-clés