| With the widespread use of satellite and remote sensing technologies,massive amounts of geographic data are being collected and stored,and traditional geological analysis cannot extract useful knowledge from the huge amount of data.Spatial data mining is a technology that combines traditional spatial data analysis methods with complex algorithms for spatial data,aiming to extract hidden knowledge from spatial data.As spatial data is usually related,meaning that the closer two spatial objects are,the more likely they have similar properties.To represent this similarity relationship,co-location patterns are proposed to represent the association between multiple objects(spatial features).As an important form of spatial data analysis,co-location patterns have been proven to guide user decisions in multiple fields,such as ecological environment,business research,and life services.To mine co-location patterns,many methods have been proposed.These methods can be divided into two categories.One category is the Apriori-like algorithms,which are based on association analysis methods for transactional datasets.They adopt a method of gradually generating higher-order co-location patterns from lower-order ones.Another category is the clique-based algorithms,which mine co-location patterns by generating maximal cliques.Both of these algorithms have high time complexity due to the need for numerous join operations or the discovery of maximal cliques in graphs.Furthermore,these two categories of algorithms only discover co-location patterns composed of fine-grained spatial features.Because they ignore users’ background knowledge,many interesting,more generalizable,and closer-to-the-essence patterns remain undiscovered.At the same time,the large number of independent co-location patterns discovered by the current framework can confuse users when making decisions.To overcome these drawbacks,this paper introduces user knowledge into the co-location pattern mining process and proposes a framework called OCPM(Co-location Pattern Miner using Ontology),which integrates ontology to guide co-location pattern mining.Moreover,this paper proposes co-location patterns composed of ontology concepts,namely Prevalent Semantic Multi-level Co-location Patterns(PSMCs).Compared with traditional patterns,PSMCs contain more interesting,generalizable,and closer-to-the-essence patterns,and PSMCs are semantically related to each other,this provides users with the convenience of quickly searching for patterns.To mine PSMCs,this paper proposes an algorithm called IDG(Instance-driven Graph),which combines the advantages of existing co-location pattern mining algorithms to achieve more efficient PSMCs mining.The OCPM framework and IDG algorithm are validated on both real and synthetic datasets,and the experimental results demonstrate their effectiveness and good performance.Furthermore,a top-down search strategy is introduced to help users quickly discover interesting patterns from PSMCs.Finally,a system called OIIKM(Ontology-based Interesting Implied Knowledge Miner)is developed,integrating the mining of PSMCs and the search for user-interest patterns. |
