A diagnosis method that matches class diagrams in a learning environment for object-oriented modeling

This paper describes the diagnostic method developed in a learning environment for object-oriented modeling called Diagram. This method compares the student's diagram with an expert's diagram in order to find the differences between these two diagrams. The principles of our algorithm rely on graph matching methods and algorithms. The diagrams are schematized in characteristic structural patterns. These structures are compared by using similarity functions, and similarity scores are computed for each couple of structures.. Finally, univalent and multivalent matches are qualified, according to a taxonomy of differences. An example is given on a real student's diagram. The algorithm is implemented in Java under Eclipse environment. It has been tested on several problems of increasing complexity, with students diagrams collected in real class situation. The results of the off-line experimentations are presented and discussed.