Relational concept analysis for relational data exploration

Abstract

Relational Concept Analysis (RCA) is an extension to the Formal Concept Analysis (FCA) which is an unsupervised classification method producing concept lattices. In addition RCA considers relations between objects from different contexts and builds a set of connected lattices. This feature makes it more intuitive to extract knowledge from relational data and gives richer results. However, data with many relations imply scalability problems and numerous results that are difficult to exploit. We propose in this article a possible adaptation of RCA to explore relations in a guided way in order to increase the performance and the pertinence of the results. We also present an application of exploratory RCA to environmental data for extracting knowledge on water quality of watercourses.