A Location Inferring Model Based on Tweets and Bilateral Follow Friends

Inferring user’s location has emerged to be a critical and interesting issue in social media field. It is a challenging problem due to the sparse geo-enabled features in social media, for example, only less than 1% of tweets are geo-tagged. This paper proposes a location inferring model for microblog users who have not geo-tagged based on their tweets content and bilateral follow friends. An approach for extracting local words from “textual” data in microblog and weighting them is used to solve the sparse geo-enabled problem, and the maximum weight location vocabulary from his/her friends or tweets is inferred as the user’s location. On a Sina-Weibo test set of 10,000 users from 10 cities, with 10 selected local words for each city, the inferring location accuracy on the city-level can reach 78.85%, and on the province-level can reach 81.39%. Compared with the TEDAS method, our method can achieve better accuracy.