Geo-tagging news stories using contextual modelling

Abstract

With the ever-increasing popularity of Location-based Services, geo-tagging a document - the process of identifying geographic locations (toponyms) in the document - has gained much attention in recent years. There have been several approaches proposed in this regard and some of them have reported to achieve higher level of accuracy. The existing geo-tagging approaches perform well at the city or country level, unfortunately, the performance degrades when the same approach is applied to geo-tag at the street/locality level for a specific city. Moreover, these geo-tagging approaches fail completely in the absence of a place mentioned in a document. In this paper, we propose an algorithm to address these two limitations by introducing a model of contexts with respect to a news story. Our algorithm evolves around the idea that a news story can be geo-tagged not only using the place(s) found in the news, but also by geo-tagging certain aspects of its context. An implementation of our proposed approach is presented and its performance is evaluated on a unique data set. Our findings suggest an improvement over existing approaches in street level geo-tagging for a specific city as well as in geo-tagging a news story even when no place is mentioned in it.