Serdyukov et al. [19] generate probabilistic language models based on the tags that photos are labeled with by Flickr users. Based on these models and Bayesian inference, they show how to estimate the location for a photo. In terms of the intention, their method is similar to our work. However, they use a GeoNames database to decide whether a usersubmitted tag is a geo-related tag, which can overlook the spatial usefulness of words that may have a strong geo-scope (e.g., earthquake, casino, and so on). Separately, the work of Crandall et al. [10] proposes an approach combining textual and visual features to place images on a map. They have restrictions in their task that their system focuses on which of ten landmarks in a given city is the scope of an image.