Using Ontological Reasoning and Planning for Data Mining Workflow Composition

This paper addresses the problem of semi-automatic design of workflows for complex knowledge discovery tasks. Assembly of optimized knowledge discovery workflows requires awareness of and extensive knowledge about the principles and mutual relations between diverse data processing and mining algorithms. We aim at alleviating this burden by automatically proposing workflows for the given type of inputs and required outputs of the discovery process. The methodology adopted in this study is to define a formal conceptualization of knowledge types and data mining algorithms and design a planning algorithm, which extracts constraints from this conceptualization for the given user’s input-output requirements. We demonstrate our approach in two use cases, one from scientific discovery in genomics and another from advanced engineering.