Spatiotemporal Pattern Mining For Users’ Access Interest On Public Map Service Platforms

Maps visually present spatial information about the world.With the development of computer technology,web mapping become popular in the form of dynamic interaction and online services to facilitate daily travel,natural resource management,and urban planning.The representation of the physical space by the web mapping constructs the web map space,relying on the Public Map Service Platforms(PMSPs).With the enrichment of spatial data,the diversification and homogeneity of spatial information leads to the reduction of effective information,and the information overload.Mining users’ access interest is the prerequisite for perceiving users’ requirements,improving the efficiency of a PMSP,and building an intelligent PMSP.The users’ interaction process in the web map space has the attributes of multiple layers’ distribution,spatial uncertainty,temporal continuity,and domains’ difference,resulting in the access interests and requirements of the users are potential and diverse,which makes achieving a proactive and efficient intelligent PMSP difficult.I achieved the semantic extraction from the explicit access contents to the implicit access interests,the patterns mining of access interests in space and time dimension,the representation of access interests from group users to domain users.The key technical problems such as the construction of the spatial mechanism and the temporal patterns mining of the access interests,and the representation of users’ access interest in different domains are solved,which promotes the multi-dimensional and multidomain perception from access interests to users’ requirements,and the development of intelligent PMSPs.From access contents in a space dimension and access operations in a time dimension,I took the PMSP trajectory as the objective,analyzed the spatial features and habitual operations,and explored the common patterns of users’ spatiotemporal access interests and spatial information demands in different domains.(1)Multi-granular access interests’ extraction based on users’ access contents.Zooming-in indicates the increase of interest and the zooming-out indicates the decrease of interest.The process of browsing behavior was defined as the change of the access interests.Modeling the PMSP trajectory and extracting its maximum access interests are our objectives.PMSP trajectory has multi-dimensional and multi-granular attributes due to the pyramid model of tiles organization making it challenging to achieve that.I constructed a space-time cube to scan the PMSP trajectory and reduce dimensionality.A new Hierarchical Gaussian Mixture Model(HGMM)was proposed to construct minimum trajectory spanning trees via recursive clustering to model the multi-granular spatial structure and extract access interests.The Random Forest model was used to improve the accuracy.This research allows us to transform the PMSP trajectories into semantic-level access interests.(2)Access interest mining for the domain users based on spatial features.I used the adjacent spatial features to represent the extracted access interests in(1)and transformed the PMSP trajectories into a collection of spatial features.I used the Word2 vec model to quantify the co-occurrence distribution of massive Points-ofInterest(POIs)to model the inherent spatial modes and constructed a multi-scale semantic space where the access interest of each user was mapped in it to improve the accuracy of the similarity measurement of access interest between users.I constructed a User-POI matrix and made multi-domains users’ classification to verify the significant difference of the access interests for the domain users(AIDUs).A combined model of the Word2 vec and singular value decomposition(W2V-SVD)was proposed to perform the AIDUs mining.(3)The representation of PMSP trajectories for multi-scale users’ access operations.To reduce the sequence error of the PMSP trajectories caused by network instability and server processing delays,the Local Dynamic Sort algorithm was proposed to perform the local reordering and sequence smoothing.I defined multi-scale operation groups composed of the relative changes of the multi-dimensional operation attributes based on the PMSP trajectory segmentation.I proposed a Browsing Behavior Representation model(BBR)to map the high-dimensional operation groups to the lowdimensional space and reduce the sensitivity of dimension and scale.Based on the common pattern mining of the access operation group,the BBR model is verified.(4)The topic PMSP trajectories mining for the domain users based on access operations.The PMSP trajectories sparsely distribute in space and vary in length.Based on the changes of the multi-dimensional operation attributes in(3)to construct the spatiotemporal invariance of the access operations and habitual characteristics.Referring to the semantic space in(2),I constructed the latent semantic space of access operations based on the BBR model in(3)to form a User-Operation matrix.I verified the significant difference of the access operations for the domain users based on the classification of domain users.The singular value decomposition(SVD)model were used to extract the topic PMSP trajectories.I achieved the mining of the spatiotemporal patterns of user access interests in different domains in the web map space,which can help PMSP providers understand the users’ spatiotemporal requirements,build user portraits,develop targeted data update strategies,reduce data production costs,and provide personalized geographic information recommendation and initiative intelligent PMSPs.