Approaches to the Selection of Relevant Concepts in the Case of Noisy Data