Nowadays, online news sources generate continuous streams of information that includes references to real locations. Linking these locations to coordinates in a map usually requires two steps involving the named entity: extraction and disambiguation. In past years, efforts have been devoted mainly to the first task. Approaches to location disambiguation include knowledge-based, map-based and data-driven methods. In this paper, we present a work in progress for location disambiguation in news documents that uses a vector-semantic representation learned from information sources that include events and geographic descriptions, in order to obtain a ranking for the possible locations. We will describe the proposed method and the results obtained so far, as well as ideas for future work.