Multi-source POI Fusion Based On Geospatial And Natural Property

With the fast development of Internet technology and the growth of Webelectronic maps, POI as the representative of geospatial data grow rapidly. POI is anabbreviation of Points of Interest, which includes title,address, latitude and longitude,classification and so on. Web POI contains a lot of value information which is a greathelp to people’s daily life and work.The POI information needs to be enriched andimproved.The traditional way to get POI information mainly relies on manual. Themethod is laborious and time-consuming and it will be replaced someday.In addition, there are differences between data, which are obtained from variouselectronic maps. And how to integrate and fuse these data from different electronicmaps, making them together, getting a richer database and realizing the effective reuseof data, finally getting structured data,which become an important problem in webdata mining and fusing.In this paper, we mainly do some research on the aspects of multi-source POIdata fusion, including choosing the feature word of POI, unifying latitude andlongitude, adding the comments, extracting the subject of the comments etc. Thespecific research and work are as follows:(1)Web data extraction and unifying latitude and longitude. First, extractinformation such as title, address, latitude, longitude reviews, phone and otherinformation of POI from some web sites, then expand the web database by thenetwork of electronic maps on title. These latitude and longitude coordinates of thesame entity are different as they are got from various electronic maps. Thus somenegative impact will be caused. POI after fusion to work caused some impact.Toresolve this problem, unifying latitude and longitude coordinates has be proposed. (2) Proposed fusion and integration of POI by analyzing the form andcharacteristics of each POI property. The main formal similarities are consists of twoparts, one part is geographic information，the other is natural property. Geographicinformation include two parts of POI, address and coordinates; the natural property isalso concludes two parts: title and comments. Address fusion is determined bycalculating the similarity of two strings.The coordinate fusion is determined bycalculating the distance between two points.Title fusion is mainly on aliasing andcomments is mainly on addition.(3) Proposed Topic Model extraction based on the understanding of the commentsand build topic model through segmentation. The experiment process of buildingmodel use some segmentation. Build a good topic model and provide effectivepretreatment for the next theme extraction from large-scale Web data.Experimental results show that the proposed technical approach can completefusion POI data automatically and efficiently, then we can get a rich Web database tofurther study.