Several Web and social media analytics require user geolocation data. Although Twitter is a powerful source for social media analytics, its user geolocation is a nontrivial task. This paper presents a purely word distribution method for Twitter user country geolocation. In particular, we focus on the frequencies of tweet nouns and their statistical matches with Google Trends world country distributions (GTN method). Several experiments were conducted, using a recently created dataset of 744,830 tweets produced by 3298 users from 54 countries and written in 48 languages. Overall, the proposed GTN approach is competitive when compared with a state-of-theart world distribution geolocation method. To reduce the number of Google Trends queries, we also tested a machine learning variant (GTN2) that is capable of matching the GTN responses with an 80% accuracy while being much faster than GTN.
Twitter User Geolocation Using Web Country Noun Searches
ZOLA, PAOLA
;Maurizio Carpita
2019-01-01
Abstract
Several Web and social media analytics require user geolocation data. Although Twitter is a powerful source for social media analytics, its user geolocation is a nontrivial task. This paper presents a purely word distribution method for Twitter user country geolocation. In particular, we focus on the frequencies of tweet nouns and their statistical matches with Google Trends world country distributions (GTN method). Several experiments were conducted, using a recently created dataset of 744,830 tweets produced by 3298 users from 54 countries and written in 48 languages. Overall, the proposed GTN approach is competitive when compared with a state-of-theart world distribution geolocation method. To reduce the number of Google Trends queries, we also tested a machine learning variant (GTN2) that is capable of matching the GTN responses with an 80% accuracy while being much faster than GTN.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
