自然邻域支持下的空间同位模式挖掘方法

空间同位模式指频繁发生在邻近空间位置的事件集合,此类模式对于深入理解不同空间要素间的交互关系具有重要意义。空间同位模式挖掘的一个核心内容是空间要素邻近关系构建,然而现有方法在空间要素分布不均匀时难以准确地描述要素间的邻近关系,容易导致挖掘结果的遗漏或误判。为此,本文提出了一种基于自然邻域的空间同位模式挖掘方法。首先从同位模式的产生机理分析入手,过滤同位模式挖掘中的干扰要素;进而,从距离邻近性、密度变化一致性和关系紧密性的原则出发,自适应地构建空间要素实例的自然邻近关系;最后,以自然邻域为基础,基于图的连通性从整体到局部发现多层次同位模式。试验分析与比较发现,本文方法能够有效发现空间要素分布不均匀情况下的同位模式,而且降低了人为设置邻域参数对挖掘结果的影响。

Discovery of co-location patterns based on natural neighborhood

Discovery of co-location patterns is crucial to understanding the interaction among different spatial features. The construction of neighborhood relationship among spatial features plays a key role in co-location pattern mining, however, existing methods are difficult to construct appropriate neighborhood relationship when the spatial features distribute unevenly.This limitation is very likely to make the omission and/or misjudgment of co-location patterns.To address this issue, a co-location pattern mining method based on natural neighborhood is proposed in this study.After removing the randomly distributed spatial features,natural neighborhood relationship among different spatial features is adaptively constructed on basis of three principles, i.e. geographic proximity, the consistency of density and compactness of neighboring relationship. The multi-level co-location patterns are discovered based on the delaunay triangulation network. The experimental results showed that the proposed method could discover the co-location patterns among unevenly distributed spatial features completely and accurately, and no user-specified parameters are required for the construction of natural neighborhood.