Abstract
Life and environmental sciences generate numerous data gathered in knowledge bases.Conceptual exploration is a data mining approach allowing to extract new pieces of knowledge. In this work, the considered conceptual exploration methods come from Formal Concept Analysis (FCA). Any method resulting from FCA imposing as input a predefined structuring of the data, the research question we address is the automatic conversion of a data or knowledge base for its exploration with FCA or with Relational Concept Analysis (RCA), an FCA extension to multi-relational data.To support this work, we used the Knomana knowledge base, which gathers descriptions of plant usages. Depending on the considered knowledge, various conversion issues must be resolved to allow relevant data mining by end users. These problems can be at the level of schemas (class model) or at the level of instances, mainly (1) converting specialization relations through inheritance hierarchy flattening; (2) converting ternary relationships through materialization or binary relations, and (3) dealing with indeterminacy of living organisms designated by the use of the abbreviation spp. in its binomial name (Linnaean).The adopted methodology is inspired by model-driven engineering, database schema transformation, software engineering refactoring and ontology conversion.This thesis presents a general conversion algorithm of a knowledge set for RCA, whose initial structure is represented in the form of a class diagram in UML format. The algorithm is applied to Knomana, and various models of the ternary relation in binary relations and a modeling of the lifting of the indeterminacy of a species name are experimented. The evaluation of this algorithm is carried out with regard to its applicability, its ability to be used for large volumes of data (i.e. scaling), and its relevance. The perspective offered by this work is to allow the analysis of any dataset and knowledge with FCA and RCA.