Spatial Information Recognition in Web Documents Using a Semi-supervised Machine Learning Method

Web documents are a promising source of spatial information. With information recognition and extraction, this information can be used in various applications such as building semantic maps and indoor robotic navigation. In this paper, we present a novel methodology to identify spatial information in web documents using semi-supervised trained machine learning classifiers. The semi-supervised models trained with the half amount of data available yield only the F-score of 4% and 9% inferior to the supervised models trained with complete data on classifying spatial entities and relationships respectively.