Where Are You Settling Down: Geo-locating Twitter Users Based on Tweets and Social Networks

摘要

In this paper, we investigate the advantages of taking two dimensions of tweet content and social relationships to construct models for predicting where people settle down as their profiles reveal city- and town-level data. Based on the users who voluntarily reveal their locations in their profiles, we propose two local word filters - Inverse Location Frequency (ILF) and Remote Words (RW) filter - to identify local words in tweets content. We also extract separately the place name mentioned in tweets using the Named Entity Recognition application and then filter them by computing the city distance. We consider users' friends and 2-hop of followings. In our experiment, we finally combine these two dimensions to estimate user location and achieve an Accuracy of 56.6% within 100 miles in city-level and 45.2% within 25 miles in town-level of their actual location which outperforms the single dimension prediction and the baseline.