They Know Where You Live!

In this paper, we demonstrate the possibility of predicting people’s hometowns by using their geotagged photos posted on Flickr website. We employ Kruskal’s algorithm to cluster photos taken by a user and predict the user’s hometown. Our results prove that using social profiles of photographers allows researchers to predict the locations of their taken photos with higher accuracies. This in return can improve the previous methods which were purely based on visual features of photos [1]. 1 Problem Definition: estimating place of living Suppose we have information about the places where a user (e.g. u) has visited during a period of time T . Let us denote every visited location like li with a triple (xi, yi, ti) where xi and yi show the latitude and longitude of the visited place and ti shows the visited time. Our problem is to predict the user u’s hometown given a sequence of visited locations with size of n. For the real data, we are going to use the available information from Flickr website . In Flickr, users have the capability to upload and share their photos with other people in the community on the website. Users can also have a profile page where they can post their personal information such as their names, hometowns, gender, occupation, list of their friends, and so on. One interesting feature of Flickr website is that it allows users to add tags to each uploaded photo. More interestingly, users can geotag their photos by explicitly posting the location that they have taken a specific photo. In this paper, we assume that users only upload the photos which have taken by themselves. Thus, the geotag information of photos can be used as a proxy for user’s mobility. www.flcikr.com