Research On The Geography Information Linking Based On Machine Learning

Geographic Information System(GIS)is comprehensive discipline which a collection of geography and cartography,computer,remote sensing technology in recent 20 years.GIS is used for input,storage,query,analysis and display the geographic data and solve the problem of complex planning and management.Now,GIS has been widely used in different fields.The integration of GIS and applied model and GIS intelligence are the key to expanding the area of GIS application.In the geographic information system,the process of collecting data for a GIS project from different sources often leads to problems of inconsistency,redundancy,ambiguity,and conflict of information in the collection,it can't contribute to the share of geographic information and the development of geographic information technology.Geospatial record linkage between entities of different geographic information sources can solve the heterogeneity of geospatial information,promote the service accuracy of the geographic information retrieval and have a great significance for geographic information integration.At present,most research works use the semantic relations,the information content and the context information to measure the similarity of geographic information,while the geospatial relationship and topology are ignored by many researches.This paper introduces approach to computing the geospatial similarity based on spatial relations,the name and categories of the entity,besides,combined with the semantic relations and machine learning method to solve geographic information link by semi-automatic mode.First of all,we collect geospatial data from three sources,namely,OpenStreetMap,Google Places and Wikimapia.We tested our approach in the metropolitan areas of both U.S.and U.K.Secondly,we build a geographic information ontology by analyzing the characteristic of geographic information,transfer data forms from a variety of data sources into a uniform benchmark before linking with mapping structured data to RDF according to the proposed ontology.Finally,To evaluate the linkage approach in support vector machine and K nearest neighbor respectively,and compared with Samal's linkage method.The results illustrated that the proposed approach obtained high correlation with human approvements.