Semantic Composition of Data Analytical Processes

Abstract

This paper presents the semantic framework for the description and automatic composition of the data analytical processes. The framework specifies how to describe goals, input data, outputs and various data operators for data pre-processing and modelling that can be applied to achieve the goals. The main contribution of this paper is the formal language for the specification of the preconditions, postconditions, inputs and outputs of the data operators. The formal description of the operators with the logical expressions allows automatic composition of operators into the complex workflows achieving the specified goals of the data analysis. The evaluation of the semantic framework was performed on the two real-world use cases from the medical domain, where the automatically generated workflow was compared with the implementation manually programmed by the data scientist.