Evaluating hypotheses in geolocation on a very large sample of Twitter

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

Recent work in geolocation has madeseveral hypotheses about what linguisticmarkers are relevant to detect where peoplewrite from. In this paper, we examinesix hypotheses against a corpus consistingof all geo-tagged tweets from theUS, or whose geo-tags could be inferred,in a 19% sample of Twitter history. Ourexperiments lend support to all six hypotheses,including that spelling variantsand hashtags are strong predictors of location.We also study what kinds of commonnouns are predictive of location aftercontrolling for named entities such as dolphinsor sharks.
Original languageEnglish
Title of host publicationProceedings of the 3rd Workshop on Noisy User-generated Text
Number of pages6
PublisherAssociation for Computational Linguistics
Publication date2017
Pages62-67
ISBN (Print)978-1-945626-94-4
Publication statusPublished - 2017
Event3rd Workshop on Noisy User-generated Text - Copenhagen, Denmark
Duration: 7 Sep 20177 Sep 2017

Conference

Conference3rd Workshop on Noisy User-generated Text
LandDenmark
ByCopenhagen
Periode07/09/201707/09/2017

Links

ID: 195014345