耦合尺度的地理实体空间相关度算法的建立与应用

传统的地理实体空间相关度算法存在适应的实体和拓扑关系类型较少、没有考虑空间尺度依赖性而导致数据区分能力差的问题。本研究提出一种能依据指定的空间尺度（本文所指“空间尺度”是指定的地理空间范围）,计算出相应的地理实体空间相关度的算法。该算法以地理学第一定律和Egenhofer关于空间相关度的论述为理论依据,分析点、线、面实体的拓扑关系和度量关系而建立不同的相关度计算公式。通过对比分析,本算法不仅能计算出不同类型和不同拓扑关系下的地理实体间相关度,而且计算结果随着空间尺度的变化而改变,与人类通常的认知相符合。最后,以地理空间数据检索为例,介绍了本算法的应用。与传统的关键词匹配检索方法相比,应用本算法能提高数据检索的F1-measure值,并且能对文档按照与检索词的相关度进行排序。本算法可应用于地理信息检索、数据发现、数据推荐和关联数据等领域。

A Case of Establishment and Application of Spatial Correlation Degree Algorithm for Geographic Entities Coupling Scales

The traditional correlation degree algorithms for geographic entities have many disadvantages, such as non-applicable for some kinds of geographic entities and some types of topological relations, and not considering the dependency of spatial scale that results in poor discernibility of data. In this study, a new algorithm is proposed which computes the spatial correlation degree according to the specified spatial scale which is represented by a spatial extent. Based on the first law of geography and the theories on spatial correlation degree put forward by Egenhofer, the equations of spatial correlation degree was obtained by analyzing the topological and metric relations between different kinds of geographical entities such as points, lines and polygons. By comparison, the algorithm in this study can compute the correlation degree between geographic entities of different types and topological relations, alter the correlation degree with the change of the specified spatial scale, which is consistent with the generic intuition of human beings. At last, we introduced an application of the algorithm by taking geospatial data retrieval as an example. Compared with the traditional retrieval methods based on keyword matching, our algorithms can improve the F1-measure in geographic information retrieval (GIR） and give the accurate scores of correlation degree so that the retrieval results can be ranked. The algorithm is an elementary research that can be applied in the research fields of GIR, scientific data discovery, data recommendation, linked data, and so on.