Research And Implementation On Key Technologies In Support- Ing Framework For Cloud-based Service-oriented Data Mining Systems

With the rapid development of cloud computing technology, data mining tech-nology and their different application areas, one cloud-based service-oriented system for data mining involves a variety of new requirements during all the analysis, design and implementation phases. In particular, explosive data in different data mining ap-plications requires distributed storage; Due to the distributed data storage, data mining algorithms need to adopt a new distributed computing model; Data mining applications are considered as complex tasks, which can not be completed by a single step rather than a flow of relevant steps; Even in the same application, different users have differ-ent requirements, so the system need to supply services according to the characteristics of the users’demands; The system involves various aspects such as connection with underlying storage and execution infrastructure, a variety of data mining algorithms, and even system performance constrains.Based on data mining applications in the cloud computing environment the overall objective of the system comprises of all subsequent phases including analysis and iden-tification of the requirements, design and scheme of the whole structure and separate components and the final implementation.This paper firstly introduces existing technologies of cloud computing, data min-ing and on-demand services. And then from the point of view of the adaptive soft-ware system, this paper proposes one supporting framework called CloudDMinerSF, abbreviated for Supporting Framework for Cloud-based Service-oriented Data Mining Systems. Three key issues are covered in CloudDMinerSF as follows.1. Adaptive requirements and overall control structure. From analysis and identifi-cation of internal and external adaptive requirements in one cloud-based service-oriented system for data mining, the type of system proves to be one self-adaptive software system. The dual closed-loop control structure and main components of CloudDMinerSF are established.2. Knowledge abstraction and strategy representation. It is necessary to introduce one knowledge base into CloudDMinerSF. And one specific description language is designed to support expression of adaptive requirements, adaptive strategies and other domain knowledge.3. Supported operating mechanisms. Based on the knowledge two types of op-erating mechanisms are supported:intelligent decision-making mechanism and scheduling mechanism. Intelligent decision-making mechanism includes rea-soning, planning, and feedback learning, and it not only satisfies the system re-quirements, but also takes efficiency into consideration. Scheduling mechanism depends on the former mechanism and mainly provides the concurrent execution for generated tasks requests as well as task action and state management.Finally, CloudDMinerSF is implemented, including the main activities such as system structure and composition, core algorithms, user interfaces and case study.