An Unsupervised Framework for Topological Relations Extraction from Geographic Documents

摘要: In this paper, we face the problem of extracting spatial relationships from geographical entities mentioned in textual documents. This is part a research project which aims at geo-referencing document contents, hence making realization Geographical Information Retrieval system possible. The driving factor huge amount Web documents mention geographic places and relate them spatially. Several approaches have been proposed for extraction relationships. However, they all assume availability either large set manually annotated or complex hand-crafted rules. both cases, rather tedious time-consuming activity required by domain experts. We propose an alternative approach based on combined use ontology, defines topological (classes) to be identified within text, nearest-prototype classifier, helps recognize instances unsupervised, so it does not need data. Moreover, prevents hand-crafting ad hoc Experimental results real datasets show viability approach.