| Humans are fast going forward into the "Big Data" era. With the growing popularity of remote sensors, location-based service terminal, spatial dataset resources are sharply increasing. Under these circumstances, service-oriented distributed Spatial Data Mining (SDM) has become an important means for gaining valuable information and discovering knowledge from distributed geospatial datasets, and further become an effectively technique for solving "Big Data" problem. While using SDM services, the integration usage of techniques, such as taking the natural language as the human-computer interaction approach, selecting the available and quality service from complex and huge SDM services collection, and automatically implementing services composition with alternative atomic services, would effectively shield the profound professional knowledge for the end service users. Especially given the recent rise of cloud service technology, it is now an important research issue of SDM and spatial knowledge discovery field.In this dissertation, the semantic analysis and intelligent solving method for SDM problem is addressed as the main study issue, the technologies of the SDM ontology construction, semantic analysis for natural language descripted SDM problem, and automatic SDM services composition based on AI planning are primarily studied, and finally accompanied with the implement of a prototype system, a solution which is able to transform the natural language descripted SDM problem into an executable services composition would be revealed.The major contents and achievements of this dissertation are concluded as follows:1. A SDM ontology construction method with core ontology building and Chinese terminology expansion is proposed firstly. SDM related concepts and terminologies are then abstracted and organized. The decimation emphasis is on the characters of spatial dataset description, such as spatial data types, spatial data services, data attribute features and spatial relation between datasets. A spatial relation determination method that adopts the administrative division as a spatial relation intermediary is presented. Then the obtained ontology concepts and relations are formalized by using OWL. A sample ontology is constructed. After that, the Chinese terminologies in the field of spatial outlier mining are extended for the SDM ontology via an ICTCLAS-based approach. Finally, the semantic register method for SDM algorithm services and spatial data services is descripted. Base on that, a semantic reasoning query method for an algorithm service and spatial data service is proposed, which is implemented by the using of Jena ontology API, SPARQL and rule-based reasoning engine.2. A novel semantic analysis method for SDM problems is presented. The roles of SDM ontology in the SDM problems semantic analysis process are confirmed firstly. Then the key semantic elements in natural language expressed SDM problem are discussed. These elements are key task and data concepts, dependency concepts and association relations. Base on the semantic elements, a triple-based semantic relation identification method is proposed. The triple-based method can effectively express the relations between semantic elements, and can easily catch and expand the semantic relations. Finally, a third-step problem semantic analysis process is descripted. It includes the processes of concept extraction, semantic pattern deduction and task formal specification. The semantic pattern deduction step is a process of appending the semantic element triple. Whiling carry out the step, the semantic triple relation detection between adjacent sentences would be ahead of longer distance sentences. The detection mechanism can fully consider semantic correlation between sentence fragments and therefore accurately grasp the semantic connotation of a SDM problem.3. For the complex multiphase character of SDM process, a multi-stage AI planning-based services composition method face to the SDM problem is proposed. Firstly, multi-stage PDDL-based planning modeling method is discussed, which is aimed to finish construction of the PDDL domain model and PDDL problem model of every stage. For the domain model, the types and predicates are converted from ontology classes and properties. Then the approach that can convert the algorithm service semantic information to the PDDL actions is descripted, which is used to fill the actions to a stage PDDL domain model. For the problem model, a method that can convert the dataset service semantic information to the PDDL problem model is presented. Throughout the works descripted above, the PDDL-based planning modeling face to different planning task and varying amounts of planning solution space is able to realize. A multi-stage services composition method based on AI planning is then detailed descripted. By the using of this method, the AI planning solution space can be effectively partitioned along with the stage divide and the services composition efficiency can be significantly improved.4. A SDM problem intelligent solving module is designed for GeoKSCloud. The key functions, such as cloud service semantic register and semantic query, problem semantic analysis and services composition based on AI planning, are implemented. With the realization of the module, a demonstration that analysis outlier of soil geochemistry data is conducted. Through which the feasibility and effectiveness of the semantic analysis and services composition intelligent planning method are confirmed. Firstly, the method can effectively shield the profound professional knowledge for the end users and reduce the thresholds of the usage of the cloud-based SDM services. Secondly, services search based on semantic query function would better meet the user needs and return more accurate and comprehensive results. Thirdly, the multi-phase AI planning method can reduce the planning time ranging from 10% to 40%, effectively improve the efficiency of automatic services composition.
|
PDF Full Text Request