Selection and Recommendation of Data Analytic Services in the Cloud

点击翻译按钮获取中文摘要

Big Data is an emerging trend and phenomenon. With more and more data becoming available from various sources, there is an increasing demand on analytic services for the purpose of understanding the data in a better way. Cloud Computing provides a platform to provision these analytic services. In this proposal, we would work on the selection and recommendation of data analytic services in the cloud. Compared to the traditional web service selection, there are several unique challenges for cloud-based analytic services which have not been sufficiently studied and which we address in this proposal; the challenges include the data-directed selection process for analytic services, the identification of meta-features appropriate for Big Data and feasible for measurement, the pairing of analytic services with other supporting cloud services (e.g., infrastructure, storage, data) to offer an end-to-end solution to users, and the ranking of the solutions. Our objective is to design and implement a data analytic service selection and recommendation system in which analytic service is selected based on the given dataset and the historical QoS values of different services on different datasets. A composite service combining analytic software service with other required component services is then recommended based on the predicted end-to-end QoS values. To begin with, we will select a problem domain such as social network analysis. For this chosen domain, we identify the meta-features of its datasets, proper data sources, and representative analytic algorithms and software services, and then we will study the problem of algorithm selection for this type of Big Data. In the next step, we will work on the service vertical composition and end-to-end QoS prediction for composite services. Finally we will build a cloud service marketplace with selection and recommendation components and evaluate our system. The key features of this proposal are as follow: 1) with the proposed selection system, users will be able to select an analytic service, which is best for the given dataset as well as the application domain, and compose a complete solution with recommended supporting cloud services; 2) we will train the HQP to gain a deep understanding on topics such as service selection and recommendation, algorithm selection and meta-learning, analytic service for Big Data, QoS prediction, and service vertical composition; in addition, they will gain hands-on experience working on system building and systematic experimental methods. Our research agenda also fits in perfectly with Canada’s Open Government initiative. Since many datasets have been made available online through this initiative, our proposed system can help ordinary citizens select the appropriate analytic services in order to create value and have a better understanding of the otherwise hardly comprehensible raw data.