Homing in on Twitter Users: Evaluating an Enhanced Geoparser for User Profile Locations

Homing in on Twitter Users: Evaluating an Enhanced Geoparser for User Profile Locations Beatrice Alex, Clare Llewellyn, Claire Grover, Jon Oberlander and Richard Tobin School of Informatics, University of Edinburgh, UK {balex|llewellyn|grover|jon|richard}@inf.ed.ac.uk Abstract Twitter-related studies often need to geo-locate Tweets or Twitter users, identifying their real-world geographic locations. As tweet-level geotagging remains rare, most prior work exploited tweet content, timezone and network information to inform geolocation, or else relied on off-the-shelf tools to geolocate users from location information in their user profiles. However, such user location metadata is not consistently structured, causing such tools to fail regularly, especially if a string contains multiple locations, or if locations are very fine-grained. We argue that user profile location (UPL) and tweet location need to be treated as distinct types of information from which differing inferences can be drawn. Here, we apply geoparsing to UPLs, and demonstrate how task performance can be improved by adapting our Edinburgh Geoparser, which was originally developed for processing English text. We present a detailed evaluation method and results, including inter-coder agreement. We demonstrate that the optimised geoparser can effectively extract and geo-reference multiple locations at different levels of granularity with an F1-score of around 0.90. We also illustrate how geoparsed UPLs can be exploited for international information trade studies and country-level sentiment analysis. Keywords: Geo-parsing, Twitter, Social Media Analytics 1. Introduction Studies taking advantage of Twitter data often need to ge- olocate Tweets or Twitter users, identifying their real-world geographic locations. Geolocation underpins analyses ranging from sentiment analysis towards particular topics or events or the tracking of disasters and epidemics, through to the mapping of breaking news stories. When analysing tweets one might want to know where users tweeted from (useful for disaster management) or where they usually live or originated from (useful for comparing groups of people’s attitudes and beliefs). Tweet-level geotag locations are en- coded as latitude/longitude (lat/long) coordinates for a very small percentage of tweets made available by Twitter (see Table 1). Many existing geography-based Twitter visuali- sations are therefore limited to this highly biased subset of data. However, tweet location is not a good proxy for a per- son’s home location and can significantly distort the results of any study or visualisation which tries to capture infor- mation for different countries or regions. By contrast, UPL is theoretically a much better approximation for home lo- cation and is much more frequently specified. In this paper, we focus on the geoparsing of UPLs to enable studies of re- gional differences between Twitter users and their attitudes. UPL is a relatively static piece of information which can be optionally specified by a user when creating their account by typing into a free form text field. The Twitter docu- mentation states that this field is: The user-defined location for this account’s profile. Not necessarily a location nor parseable. The data is entered in textual form, and places are not geolocated. Hecht et al. (2011) present an analysis of Twitter UPLs. They show that in their data …

Content truncated to 500 words