Spatial pattern matching with quantitative and qualitative constraints.

Abstract

Geo-textual searches involve keywords and spatial location restrictions. One example is the search for Points of Interest (POIs), such as schools and supermarkets, in applications such as Google Maps. Notably, most systems perform separate searches for each type of POI. Recent studies have proposed mechanisms to retrieve groups of geo-textual heterogeneous objects, closely located and relevant to a set of keywords. The Spatial Pattern Matching (SPM) query retrieves groups of POIs or other geo-textual objects based on spatial patterns with keywords and distance thresholds, although it does not consider qualitative requirements such as connectivity between objects. Consequently, SPM algorithms cannot efficiently solve queries such as “finding shopping malls that contain a training gym inside”. In this sense, this dissertation investigates “Quantitative and Qualitative Spatial Pattern Matching” (QQ-SPM), a more flexible type of geo-textual search with keywords, distance, topological and exclusion constraints between the searched geo-textual objects. A mathematical formalization and an efficient approach composed of three solution strategies for QQ-SPM searches are proposed in this research. The first proposed solution, QQESPM-Quadtree, is independent of spatial databases and uses on-disk IL-Quadtree indexing. The second, QQESPM-Elastic, converts the spatial pattern of the search into native spatial Elasticsearch queries. The third, QQESPM-SQL, transforms the spatio-textual search requirements into a single and efficient SQL query, employing spatial functions and indexing in PostgreSQL. Experiments using a dataset of POIs from London compared the effectiveness and efficiency of the three proposed solutions for QQ-SPM queries. The results confirmed the effectiveness of the proposed formalization and approach. The QQESPM-SQL solution excelled in scalability by presenting robust execution times for larger datasets. However QQESPM-Quadtree and QQESPM-Elastic presented advantages for some specific search scenarios.