Graph-based spatial co-location pattern mining: integrate geospatial analysis and logical reasoning

• Published in: International journal of digital earth Vol. 17; no. 1
• Main Authors: Jinghan Wang, Tinghua Ai, Hao Wu, Haijiang Xu, Tianyuan Xiao, Guangyue Li
• Format: Journal Article
• Language: English
• Published: Taylor & Francis Group 31.12.2024
• Subjects: adaptive neighborhood graph; graph growth; Neo4j database; POI; Spatial co-location pattern mining
• ISSN: 1753-8947; 1753-8955
• DOI: 10.1080/17538947.2024.2390434

Spatial co-location patterns reflect the inherent correlations among geographical elements. Mining co-location patterns of POIs can provide valuable insights for urban planning and resource management. Generally, co-location mining comprises two steps: proximity relationship determination (geospatial analysis) and frequent pattern recursion (logical reasoning). Previous methods often separate these two steps: serializing proximity relationships to enumerate frequent sequences. However, this approach suffers from limited flexibility and intuitiveness: as continuous spatial contexts are segmented into numerous small parts, it fails to adequately represent geographic correlations and hinders the effective visualization of logical reasoning. Facing these challenges, this study proposes a novel graph-based spatial co-location mining method (GSCM), which leverages graphs to integrate geospatial analysis and logical reasoning. Initially, to establish adjacency relationships, GSCM constructs the adaptive neighborhood graph, which dynamically adjusts proximity thresholds to accommodate geographic heterogeneity. Subsequently, the Apriori logical recursive process is realized on the graph structure. By leveraging graph searching, pruning, and growing, the potential growth directions of co-location patterns are identified, enhancing both the efficiency and intuition of frequent pattern recursion. Through experiments conducted on large-scale POI datasets from Wuhan, GSCM is compared with existing baseline methods, verifying its potential to uncover co-location patterns in complex spatial contexts.