Classifying natural-language spatial relation terms with random forest algorithm

摘要: The exponential growth of natural language text data in social media has contributed a rich source for geographic information. However, incorporating such GIS analysis faces tremendous challenges as existing tend to be geometry based while rely on spatial relation NLSR terms. To alleviate this problem, one critical step is translate geometric configurations into terms, but methods date e.g. mean value or decision tree algorithm are insufficient obtain precise translation. This study addresses issue by adopting the random forest RF automatically learn robust mapping model from large number samples and evaluate importance each variable term. Because semantic similarity collected terms reduces classification accuracy, different grouping schemes used, with their influences results being evaluated. experiment demonstrate that learned can accurately transform recognizing groups require sets variables. More importantly, evaluation indicate topology types determined 9-intersection weaker than metric variables defining which contrasts assertion ‘topology matters, refines’ studies.