The Research On The Semantic Based Building Address Standardization

Smart cities can effectively use information and communication technologies to achieve sensing,analysis,and integration of key information of urban operating systems,so as to respond intelligently to various needs including public safety,urban services,and building energy conservation,creating for humanity.A wonderful city life experience.However,for different urban operating systems,the data representation and transmission technology used are different.If you want to integrate the intelligent system data of multiple building units,break the “information island” and achieve high-quality data fusion,it is urgent to find connectivity.A valid 'medium' for class data.As a universally recognized architectural orientation representation carrier,the construction address is a unique and fixed coding method for each building because of its good address information expression form.It is a good medium for data fusion of intelligent building systems.Combined with geocoding technology,it can also realize the map of the building address;by searching,it can quickly locate the building location on the map,extract the building body data,provide information technology support for various system optimizations,but the building address also has its own problems,such as The Chinese address planning started late and did not use a uniform address standard;the address standards adopted by each institution were different;the Chinese address was freely expressed,the matching workload was large,and the matching difficulty was high,so the address representation standard was unified,and the standard address and multi-agency address were realized.The precise matching between them is an urgent problem to be solved.In view of the above problems,this paper trains the address matching model based on convolutional neural network based on the address data provided by a public security system and other energy supply systems to achieve fast and accurate matching of public security standard addresses and other energy supply unit addresses.Due to the large number of building addresses and semantic differences,methods such as exhaustive matching cannot be effectively applied.Therefore,this paper firstly washes the address data quickly,finds ten corresponding standard candidate addresses for each query address,and narrows the matching problem from hundreds of thousands of candidate spaces to ten data items.Subsequently,for the training neural network model In order to solve the problem of standard training data,this paper uses crowdsourcing annotation technology to decompose and publish data annotation tasks to multiple labelers to quickly obtain high-quality annotation data.Then,by analyzing the general rules and hierarchical structure of building address semantics,The address is carefully segmented.Finally,based on the DSSM network design and implementation of a one-to-many address matching system,83% matching accuracy is achieved.In order to realize the visualization of address matching,this paper also combines Baidu map API to realize the address map.