Leveraging local interactions for geolocating social media users
2018
Online
Elektronische Ressource
© Springer International Publishing AG, part of Springer Nature 2018. Predicting the geolocation of social media users is one of the core tasks in many applications, such as rapid disaster response, targeted advertisement, and recommending local events. In this paper, we introduce a new approach for user geolocation that unifies users’ social relationships, textual content, and metadata. Our two key contributions are as follows: (1) We leverage semantic similarity between users’ posts to predict their geographic proximity. To achieve this, we train and utilize a powerful word embedding model over millions of tweets. (2) To deal with isolated users in the social graph, we utilize a stacking-based learning approach to predict users’ locations based on their tweets’ textual content and metadata. Evaluation on three standard Twitter benchmark datasets shows that our approach outperforms state-of-the-art user geolocation methods.
Titel: |
Leveraging local interactions for geolocating social media users
|
---|---|
Link: | |
Veröffentlichung: | 2018 |
Medientyp: | Elektronische Ressource |
Schlagwort: |
|
Sonstiges: |
|