Text-based ontology construction using relational concept analysis

International audienceWe present a semi-automated process that constructs an ontology based on a collection of document abstracts for a given domain. The proposed process relies on the formal concept analysis (\fca), an algebraic method for the derivation of a conceptual hierarchy, namely '\textit{concept lattice}', starting from data context, i.e., set of individuals provided with their properties. First, we show how various contexts are extracted and then how concepts of the corresponding lattices are turned into ontological concepts. In order to refine the obtained ontology with transversal relations, the links between individuals that appear in the text are considered by the means of a richer data format. Indeed, Relational Concept Analysis (\rca), a framework that helps \fca in mining relational data is used to model these links and then inferring relations between formal concepts whose semantic is similar to roles between concepts in ontologies. The process describes how the final ontology is mapped to logical formulae which can be expressed in the Description Logics (\dl) language ${\cal FLE}$. To illustrate the process, the construction of a sample ontology on the astronomical field is considered