“Formal Concept Analysis: Themes and Variations for Knowledge Discovery and Knowledge Representation”

In this presentation, firstly we will introduce the main principles of Knowledge Discovery guided by Domain Knowledge (KDDK as it is carried out in the Orpailleur Research Team at LORIA). A Knowledge Discovery (KDD) process is applied to data which are prepared for being the mined. The extracted patterns are the validated and interpreted (by an analyst) for being reused within a knowledge system. In addition, the KDD process can be qualitatively enhanced when domain knowledge is used at each step of the process. Among KDD methods, Formal Concept Analysis (FCA) holds a particular place: as a classification method, FCA can be used for both extracting and representing knowledge with different points of view. We will present two main variations of FCA, namely Relational Concept Analysis (RCA) and Pattern Structures, for extending the standard framework of FCA. We will give working examples of these two extensions, especially in text mining and in classification.