Data Service API Design for Data Analytics.

Data service APIs provide uniform and filtered interfaces for data analysts to retrieve data. However, existing RESTful data services do not serve data analytics well because most of them are designed based on the underlying data schema rather than aligning with the requirements of data analytics. First, the API representations only support one-off communication, which lacks analytic semantics to guide analysts to continuously explore and retrieve data. Second, the current data service design does not support re-usage of data exploration processes and derived data generated from data analysts. In this paper, we propose an analytics-focused API design for data services. First, we introduce a service architecture and its resource APIs to realize core functions of data retrieval. Second, we design a navigation model for analysts to navigate resource APIs more efficiently. Third, we extend and leverage data package technique to provide context information about the origin, scope, and historical manipulations on a certain dataset. This mechanism allows the analysts to share and reuse historical data exploration process and derived data. We evaluate our approach using a case study and compare our approach against the conventional data APIs. The evaluation shows that our approach has advantages over traditional data service APIs in maturity, interoperability, discoverability, and reusability.