大数据环境下基于贝叶斯推理的中文地名地址匹配方法

摘要

传统的中文地名地址匹配技术难以处理大数据环境下海量、多样和异构的智慧城市地理信息空间中的中文地名地址快速匹配问题.提出了一种Spark计算平台下基于中文地名地址要素的匹配框架及应用智能决策的匹配算法(An Intelligent Decision Matching Algorithm,AIDMA).首先,从中文地名地址中富含的语义性和中文字符串、数字与字母之间的自然分隔性两个方面进行地址要素解析,构建了融合多距离信息的贝叶斯推理网络,从而提出了基于多准则评判的中文地名地址匹配决策方法.然后,利用芜湖市514967条脱敏后的燃气开户中文地名地址信息库与1770979条网格化社区中的中文地名地址信息库(包含网格化地址的地理空间信息)进行实验与分析.实验结果表明,在处理大规模中文地名地址信息时,相比于传统的中文地名地址匹配方法,该方法能够有效提高单条中文地名地址的匹配效率,同时在匹配度与精确度两个指标上匹配结果更加均衡.%Traditional matching technologies of Chinese place-name address is hard to deal with the fast matching problem of Chinese place-name address in matching massive,diverse and heterogeneous geographic information under the big data environment.An intelligent decision matching algorithm(AIDMA) based on computing framework of Spark was proposed.Firstly,geographical elements are analyzed from semantic information and separations of Chinese strings,numbers and letters.Bayesian networks is constructed with three kind of distance combined with multi-criteria decisionmaking effectively.514957 desensitized gas account information and 1770979 grid addresses information which includes spatial information of Wuhu City are used to perform the experiments.The conclusions prove that the executed time of each record of AIDMA is reduced to about 2.2s from 1min when compared to traditional algorithms.The matching results are more balanced on matching rate and precise rate.The proposed method possesses the theoretical significance and application value on the road to construct the intelligent countries.