Approaches to estimate location of social network users based on social graph

Many applications require information about the geolocation of users, which is not always available. Among the users of Twitter only about 26% indicate the name of the city in their profiles, about 30% of users of VKontakte leave this field blank. So there is the problem of determining the place of residence of social network users. We investigate approaches to geolocation of social network users using their mutual bidirectional ties - social graph. At first, we present a brief overview of the work in the field of geolocating users of social networks. Then we propose an approach that relies on graph nodes' embeddings and supervised machine learning techniques. Series of experiments were conducted with proposed and baseline approaches. Experiments show that proposed approach is comparable with others. The results of experiments allow us to conclude that the proposed approach based on vector representation can be effectively used to determine the user's place of residence by itself, or in combination with classifiers based on user data It is worth noting that the proposed approach has no any specifics related to the geolocation. It can also be used to assess any other demographic attributes that influence the formation of relationships in society. Thus, a similar approach was used in Perozzi and Skiena to determine the age of the users.