Mining Prevalent Co-location Patterns Utilizing Graph Database And Fuzzy Technologies

A spatial prevalent co-location pattern is a subset of a set of spatial features whose objects frequently occur together in geographic space.Spatial co-location pattern mining aims to extract unknown but potentially useful information from spatial data to better serve human activities,and it has driven many social applications,such as location-based services,urban planning,etc.Although prevalent co-location pattern mining has carried out many explorations,there are still some problems:(1)the calculated spatial proximity relationship is usually stored in memory,which is more efficient to collect table instances,but as the amount of data increases,it will occur more memory and store objects repeatedly;(2)many existing methods treat the space as homogeneous,use absolute Euclidean distance to measure the neighbor relationship between objects,ignore the proximity between objects,and the distance between objects is a relative and fuzzy concepts that cannot be precisely defined;(3)the aggregate distribution of spatial objects leads to a complex sharing relationship between objects,and the use of participation index to measure the prevalence of patterns ignores this relationship,making the mined patterns have low practical significance;(4)collecting participating objects by generating table instances of candidate schemas has large time and space overhead,resulting in low computational efficiency of the algorithms.To deal with the above problems,this thesis first explores a spatial prevalent colocation pattern mining method based on graph database technology.The graph database stores data in the native graph structure,which can materialize spatial data and proximity relationships well.Based on the materialized neighborhood graph,mining algorithms based on subgraph search(CliqueSearch)and central object filtering and verification(ObjectSearch)are designed respectively,and the correctness and completeness of the proposed algorithms are also proved.Secondly,a mining method based on fuzzy technology is proposed,i.e.,spatial co-location patterns mining utilizing fuzzy grid cliques.Specifically,the fuzzy theory is introduced to define the proximity between objects;considering the sharing relationship and proximity between objects,the fuzzy participation contribution index is defined to measure the interest of the patterns;based on the defined proximity metrics,a basic mining framework utilizing fuzzy grid cliques is proposed.Then,based on the proposed mining framework,a naive algorithm called participating object filtering and verification(POFV)is designed,which uses fuzzy grid clique search technique instead of combined search to collect participant objects and avoids enumerating all table instances.To solve the bottleneck of the naive algorithm,an algorithm called maximal fuzzy grid clique search for participating objects(MFGC)was developed,which can reuse information efficiently.Finally,experiments on real and synthetic datasets demonstrate that the proposed CliqueSearch and ObjectSearch algorithms based on graph database technology,POFV,and MFGC algorithms based on fuzzy technology have better performance in mining results and execution performance.