Detecting geographic locations from web resources

Abstract

The rapid pervasion of the web into users' daily lives has put much importance on capturing location-specific information on the web, due to the fact that most human activities occur locally around where a user is located. This is especially true in the increasingly popular mobile and local search environments. Thus, how to correctly and effectively detect geographic locations from web resources has become a key challenge to location-based web applications. In our previous work, we proposed to explicitly distinguish three types of locations for web resources, namely provider location, content location and serving location. Provider location is the physical location of the provider who owns the web resource; content location is the geographic location described in the web content; while serving location is the geographic scope that a web resource can reach. In this paper, we present a system that comprehensively employs a set of algorithms and different geographic sources by extracting geographic information from the web content, and mining hyperlink structures as well as user logs. As the result, only relevant geographic sources, rather than all of possible ones are used in computation of each category of web location. Finally, experimental results on large samples of web data show that our solution outperforms previous approaches.