Supervised Text-based Geolocation Using Language Models on an Adaptive Grid

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 1500–1510, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Supervised Text-based Geolocation Using Language Models on an Adaptive Grid Stephen Roller† Michael Speriosu ‡ Sarat Rallapalli † Benjamin Wing ‡ Jason Baldridge ‡ †Department of Computer Science, University of Texas at Austin ‡Department of Linguistics, University of Texas at Austin {roller, sarat}@cs.utexas.edu, {speriosu, jbaldrid}@utexas.edu, ben@benwing.com Abstract The geographical properties of words have re- cently begun to be exploited for geolocating documents based solely on their text, often in the context of social media and online content. One common approach for geolocating texts is rooted in information retrieval. Given training documents labeled with latitude/longitude co- ordinates, a grid is overlaid on the Earth and pseudo-documents constructed by concatenat- ing the documents within a given grid cell; then a location for a test document is chosen based on the most similar pseudo-document. Uniform grids are normally used, but they are sensitive to the dispersion of documents over the earth. We define an alternative grid con- struction using k-d trees that more robustly adapts to data, especially with larger training sets. We also provide a better way of choosing the locations for pseudo-documents. We eval- uate these strategies on existing Wikipedia and Twitter corpora, as well as a new, larger Twit- ter corpus. The adaptive grid achieves com- petitive results with a uniform grid on small training sets and outperforms it on the large Twitter corpus. The two grid constructions can also be combined to produce consistently strong results across all training sets. 1 Introduction The growth of the Internet in recent years has provided unparalleled access to informational re- sources. It is often desirable to extract summary metadata from such resources, such as the date of writing or the location of the author – yet only a small portion of available documents are explicitly annotated in this fashion. With sufficient training data, however, it is often possible to infer this infor- mation directly from a document’s text. For exam- ple, clues to the geographic location of a document may come from a variety of word features, e.g. to- ponyms (Toronto), geographic features (mountain), culturally local features (hockey), and stylistic or di- alectical differences (cool vs. kewl vs. kool). This article focuses on text-based document ge- olocation, the prediction of the latitude and lon- gitude of a document. Among the uses for this are region-based search engines; tracing the sources of historical documents; location attribution while summarizing large documents; tailoring of ads while browsing; phishing detection when a user account is accessed from an unexpected location; and “activist mapping” (Cobarrubias, 2009), as in the Ushahidi project.1 Geolocation has also been used as a fea- ture in automatic news story identification systems (Sankaranarayanan et al., 2009). One of the first works on document geolocation is Ding et al. (2000), who attempt to automatically de- termine the geographic scope of web pages. They …

Content truncated to 500 words