%0 Journal Article
%T 基于Voronoi图和距离衰减效应的模糊实例空间并置模式挖掘算法<br>Fuzzy Instance Space Co-Location Pattern Mining Algorithm Based on Voronoi Graph and Distance Attenuation Effect
%A 陈书健
%A 芦俊丽
%J Hans Journal of Data Mining
%P 65-80
%@ 2163-1468
%D 2024
%I Hans Publishing
%R 10.12677/hjdm.2024.142006
%X 空间并置模式挖掘用于发现一组空间特征，它们的实例在空间中频繁地相互邻近。传统的空间并置模式挖掘过程中，将空间实例抽象成点对象，每个实例对应一个确定位置。然而，规模较大的空间实例有多个重要位置点(如医院、公园入口)，其空间位置因对其入口的认知不同而存在差异，具有模糊性。对于这些模糊实例，本文考虑其重要位置点对该实例规模的贡献，重新定义实例间的邻近度。此外，传统的并置模式挖掘方法忽略了特征实例的空间分布密度以及邻近实例间的邻近程度，采用静态的距离阈值来识别邻近实例。本文考虑特征的分布密度，用Voronoi图自适应提取不同特征的邻近实例，结合邻近实例的距离衰减函数，更加科学地描述实例间的邻近度。提出一种同时考虑模糊实例规模和距离衰减效应的空间并置模式挖掘方法，为实现快速挖掘，设计了极大团和哈希表搜索参与实例的挖掘框架。在真实数据集和合成数据集上进行实验，验证本文的算法可以发现传统空间并置模式挖掘方法所忽略的有意义模式。<br />
Space co-location pattern mining is used to discover a set of spatial features whose instances are frequently adjacent to each other in space. In the process of traditional space co-location pattern mining, spatial instances are abstracted into point objects, and each instance corresponds to a definite location. However, large spatial examples have multiple important location points (such as hospital, and park entrance), and their spatial locations are different due to different cognition of their entrances, which is fuzzy. For these fuzzy instances, this paper considers the contribution of important location points to the instance scale, and redefines the proximity between the instances. In addition, the traditional collocation pattern mining method ignores the spatial distribution density of feature instances and the proximity degree between neighboring instances, and uses static distance threshold to identify neighboring instances. In this paper, considering the distribution density of features, Voronoi diagram is used to adaptively extract adjacent instances with different features, and the proximity of adjacent instances is described more scientifically by combining the distance decay function of adjacent instances. In this paper, a space co-location pattern mining method is proposed, which takes into account both the fuzzy instance size and distance attenuation effects. In order to realize fast mining, a mining framework for the participating instances of maximal clique and hash table search is designed. Experiments are carried out on real data sets and synthetic data sets to verify that the proposed algorithm can find meaningful patterns ignored by traditional space co-location pattern mining methods.
%K 空间并置模式，模糊实例，距离衰减效应，Voronoi图<br>Space Co-Location Pattern
%K Fuzzy Instance
%K Distance Attenuation Effect
%K Voronoi Diagram
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=84073