Geocoding textual documents through the usage of hierarchical classifiers

摘要: In this paper, we evaluate automated techniques, based on a hierarchical representation for the Earth's surface and leveraging SVM classifiers, assigning geospatial coordinates to previously unseen documents, using only raw text as input evidence. We report experiments with Wikipedia documents in four different languages, two Twitter datasets from previous studies. obtained state-of-the-art results, showing that document geocoding can be handled effectively appropriate bag-of-words representations out-of-the-box supervised learning methods.