UPDATE: We were quoted in this related Mashable article: How Political Campaigns Can Turn Social Media Support Into Votes
The holy grail of politics on Twitter has been identifying the geographic location of users. We get asked for a solution to this problem all the time. Unfortunately, it isn’t as easy as it seems. There are two official ways of finding a user’s location, and neither of them work. The location field in the user profile is rarely filled in with a real-world location, like a US city and state. Most of the time people show their wit through common ironies, like “here”, “there”, or “everywhere.” The other failure point is geocoding tweets. While Twitter allows you to let your mobile phone capture your latitude and longitude when you tweet, it isn’t used. Out of close to 9 million tweets we have collected for the 2012 election, less than 0.5% include a geocode.
Finally, we have the solution. We call it contextual geocoding. If a user repeatedly tweets about something that is fixed in space, or follows an account associated with a geographical region, we can assume they are interested in that location. This technique is taking shape in our database for 2012twit.com. A month ago we gave up on Sarah Palin (Run, Sarah, run, it’s not too late!), and started collecting tweets that mention all incumbent congressmen, senators, and governors. We soon realized that this data could be used to locate Twitter users geographically. Each incumbent politician is located in a specific state. The more a user tweets about politicians from a state, the greater their connection to that location.
Now we’ve started making that connection even stronger by collecting the lists of all followers of all incumbent politicians. The more state politicians someone follows, the more they show their interest in that state. By cross referencing a user’s follows with their tweets, we can identify users who are strongly associated with one or more states. How much data is there? An amazing amount. In one experiment, we collected the followers of all incumbent Democrats, and found 689,000 unique Twitter accounts. This is shaping up to be a great lead list for Twitter voters organized by location.