融合神经网络和空间关系的中文地址解析  

Chinese address understanding by integrating neural network and spatial relationship

作　　者：刘现印[1] 李玉琳[1] 尹斌[1] 田沁 LIU Xianyin;LI Yulin;YIN Bin;TIAN Qin(Shandong Provincial Land Surveying and Mapping Institute,Jinan 250013,China;Shenzhen Research Center of Digital City Engineering,Shenzhen,Guangdong 518034,China;Key Laboratory of Urban Land Resource Monitoring and Simulation,Ministry of Natural Resource,Shenzhen,Guangdong 518034,China)

机构地区：[1]山东省国土测绘院,济南250013 [2]深圳市数字城市工程研究中心,广东深圳518034 [3]自然资源部城市国土资源监测与仿真重点实验室,广东深圳518034

出　　处：《测绘科学》2021年第8期165-171,212,共8页Science of Surveying and Mapping

基　　金：山东省重大科技创新工程项目(2019JZZY020103);山东省“十三五”基础测绘规划基金资助项目(201605097)。

摘　　要：中文地址解析是中文地址匹配最核心的问题。针对当前比较流行的基于条件随机场(CRF)或者基于规则的中文地址解析方法,该文结合深度学习中的双向门控循环网络(BiGRU)和CRF的方法来实现中文地址分词;并且针对当前的层次地址模型和四词位标注体系,该文采用了基于空间关系地址模型和五词位的标注方法。然后分别采用基于规则的模型、CRF、BiGRU+SoftMax和BiGRU+CRF模型进行对比实验,发现该文提出的BiGRU+CRF模型配上新的空间关系地址模型及标注体系,可以对地址解析方面有更好的效果。Chinese address understanding is the most difficult problems in Chinese geocoding. Currently, most methods are rule-based or CRF-based models. In the paper, we try to use BiGRU+CRF model to solve this problem. In addition, we improve the 4-tags labeling architecture with 5-tags labeling architecture, in the meanwhile, we use spatial relationship address model to replace the most used hierarchical address model. At last, we conduct some experiments to compare the performance of rule-based model, CRF,BiGRU+SoftMax and our model, we find that our model has achieved the best result.

关 键 词：空间关系地址模型 中文地址解析 深度学习 双向门控循环网络(BiGRU) 

分 类 号：P208[天文地球—地图制图学与地理信息工程]