Monitoring users' emotive states and using that information for providing feedback and scaffolding is crucial. In the learning context, emotions can be used to increase students' attention as well as to improve memory and reasoning. In this context, tutors should be prepared to create affective learning situations and encourage collaborative knowledge construction as well as identify those students' feelings which hinder learning process. In this paper, we propose a novel approach to label affective behavior in educational discourse based on fuzzy logic, which enables a human or virtual tutor to capture students' emotions, make students aware of their own emotions, assess these emotions and provide appropriate affective feedback. To that end, we propose a fuzzy classifier that provides a priori qualitative assessment and fuzzy qualifiers bound to the amounts such as few, regular, and many assigned by an affective dictionary to every word. The advantage of the statistical approach is to reduce the classical pollution problem of training and analyzing the scenario using the same dataset. Our approach has been tested in a real online learning environment and proved to have a very positive influence on students' learning performance.