Research On Self-learning Construction Method Of Chinese Address Element Library Based On Internet POI

With the development of Internet technology,more and more spatial location information is involved in Web pages,making it one of the important data sources for geographic information data acquisition and update.However,due to the diversity,complexity and heterogeneity of Internet data,their mining and analysis has become a new problem.The research on the construction method of Chinese address element database is one of the problems.As the core hub of information resource integration,fusion and management,the address element library undertakes the functions of information resource conformity and supporting technology application,providing reliable,efficient and accurate geographical location services for the public and government functional departments.According to the description characteristics of Chinese address text information and the requirement of address element matching service,this paper systematically studies Chinese address element segmentation,semantic annotation and hierarchical relationship construction,and designs and develops a prototype system.The main research work and innovations are as follows:(1)Research on Chinese address element segmentation method based on GRU:Aiming at the shortcomings of traditional machine learning model in segmentation of Chinese address elements,artificial feature extraction and long training time are needed.A Chinese address element segmentation method based on Gated Recurrent Unit(GRU)neural network is proposed,and three-word position labeling method is used to label the segmented characters.The neural network has the feature of automatic learning,which avoids the influence of artificial feature selection on Chinese address segmentation.At the same time,the three-word position tagging method improves the performance of Chinese address element segmentation and reduces the training time of the model.Based on various types of catalogue data and POI interest point address data in Baidu Map,the training and validation experiments of the neural network model are carried out.The results show that compared with the traditional machine learning model,the network model has a significant improvement in segmentation performance and model efficiency.(2)Semantic annotation and hierarchical relationship construction of address elements:Semantic annotation of Chinese address elements usually uses the semantic relationship between address texts to annotate,thus ignoring the shortcomings of spatial location information of address elements.In this paper,we propose a semantic annotation method for address elements based on keyword matching and location inference,and construct the hierarchical relationship between address elements through Tire tree.The experimental results show that this method has obvious advantages over other methods in terms of labeling efficiency,accuracy and coverage.(3)Design and implementation of prototype system:According to the relationship among the Chinese address element tables,this paper designs seven database tables with different structures.On this basis,a Chinese address element management system is developed,which achieves the basic operations of segmentation of Chinese address elements,statistics of Chinese address expression patterns,and addition,deletion,modification and search of Chinese address elements.