Study On Fusion Of Multi-source POI Based On The Spatial Location Information

With the development of Web Electronic Maps and Location Based Services(LBS), the geospatial data such as POI appeared to grow rapidly. POI is theabbreviation of the points of interest (Point of interest), a point data which representsthe true geographic entity, generally, POI includes basic information such as name,category, latitude, longitude as well as address. On the one hand, the collection,storage and update of POI information take lots of manpower and material,meanwhile, addition and update services of the POI information have become the corecompetitiveness of various network maps; on the other hand, the POI informationfrom different sources have different maturity level. How to integrate and fuse thePOI information from different sources has become an urgent problem.The fusion technology of POI information is a key technology to solve thisproblem. The ultimate goal of the POI data fusion technology mentioned in this articleis: identity the POI which represents the same geographical entity from two POIdatasets, and put them into fusion set finally. So far, foreign researchers proposedsome solutions such as Ontology-based technology; technology based on spatiallocation; technology based on non-spatial attributes. To find out the correspondingobjects from distinct original POI datasets, we propose a novel approach on the basesof the abroad researching results. This approach increases the accuracy of the fusionset by using non-spatial properties similarity based on the spatial location. Thedetailed research works and results are as follows:(1) Apply location-based algorithms to two POI datasets to find out initial fusion setconsisting of corresponding objects. The advantage of the location-basedalgorithms is that you can fine the corresponding objects just use the latitude andlongitude attributions and each POI almost have the latitude and longitudeattributions, however, the shortcoming is that the latitude and longitude of POIfrom different sources have prevalent error as well as the not unified coordinatesystem.(2) Use the methods based on name information with low threshold to exclude wrong corresponding objects obtained using location-based method. The advantage ofthis algorithm is that it uses only non-spatial characteristics without regard todifferences in latitude and longitude, such method is also more mature, thedisadvantage is that it requires different sources POI must have unified storagemode, additionally there maybe a lack of information and tagging errors in thenon-spatial characteristic attributes. The reason to use the methods based on nameinformation with low threshold is that the POI having similar name will havesimilar name.(3) The remain corresponding objects not found will be searched out by using themethods based on name information with high threshold. The reason to use themethods based on name information with high threshold is that the secondaryfilter of the POI data is not the spatial location of the constraint.Finally, this modified technique was tested on the different POI datasets. Ithas been demonstrated by our experiments that the precision, recall and F1-valueof the fusion set was improved to a large part.