顾及环境相似的多特征组合实体匹配方法

Entity Matching Methods Based on Combining Multi-Similarity-Characteristics Considering Environment Similarity

实体匹配是指通过相似性指标识别出不同来源地图数据中同名实体并建立实体对应关系的过程,它是地图数据融合与更新中的关键技术和难点之一,匹配效果的好坏直接关系到后续应用的正确性。现有的实体匹配方法大多只利用实体自身的几何或属性特征进行匹配,很少顾及实体周边环境的相似性,使得一些匹配实例不足以得到正确的匹配结果。针对该点不足,首次引入点、线实体环境相似度,并提出顾及环境相似的多特征组合匹配方法,有效地提高了实体匹配精度;通过双向匹配与聚类合并策略,解决了部分线实体一对多、多对多的匹配问题。

The entity′s geometric and attribute characteristics are used in existing matching methods,and the environment similarity of entities is seldom considered,it leads to wrong matching result for some matching instances.According to this deficiency,the environment similarity measures for point and line entity are brought forward,where point entity environment similarity measure is formed as follows：Firstly,2×2 grid around entity is constructed and its environment characteristic is calculated,which assumes that four grids are described respectively by LT,RT,RB and LB,then environment characteristics of point entity can be described by four-unit group（ρLT,ρRT,ρRB,ρLB）,ρ denotes point density of a grid region.Secondly,the size of corresponding items are compared in four-unit group for matching entities,if their sizes are equal,the count variable will be added 1,otherwise count variable is unchanged.Lastly,environment similarity measure is equal to 1/4 of count variable.Environment similarity measure of line entity is determined by the similarity degree of adjoining graph structure of line entities,adjoining graph is composed of current entity and all entities that directly connect to current entity.The structure relationship between current entity and it′s adjoining entity is calculated,and the structure relationship can be described with the dual group（distance,angle）,where distance denotes the normalized path length from connection point to start point of current line entity,angle denotes normalized azimuth of line segment in adjoining entity which connects to current entity.Then all structure relationships of matching pairs are matched.Last,the environment similarity measure is calculated according to matching result of structure relationships.Based on introducing environment similarity measure,matching methods are put forward based on combining multi-similarity-characteristics considering environment similarity,it improves effectively the entity matching precision,through the matching strategy of bidirectional matching and clustering combination,and resolves some entity matching problems of one-to-many and many-to-many.