Semantic subgroup explanations

摘要

Subgroup discovery (SD) methods can be used to find interesting subsets of objects of a given class. While subgroup describing rules are themselves good explanations of the subgroups, domain ontologies can provide additional descriptions to data and alternative explanations of the constructed rules. Such explanations in terms of higher level ontology concepts have the potential of providing new insights into the domain of investigation. We show that this additional explanatory power can be ensured by using recently developed semantic SD methods. We present a new approach to explaining subgroups through ontologies and demonstrate its utility on a motivational use case and on a gene expression profiling use case where groups of patients, identified through SD in terms of gene expression, are further explained through concepts from the Gene Ontology and KEGG orthology. We qualitatively compare the methodology with the supporting factors technique for characterizing subgroups. The developed tools are implemented within a new browser-based data mining platform ClowdFlows.