Automated Class Diagram Assessment using Semantic and Structural Similarities

Class diagrams show classes in software and the relationships between those classes. A class diagram is a unified modeling language diagram commonly used in education. Thus, an assessment of class diagrams is essential for teachers who usually have students produce class diagrams based on predetermined projects. Teachers assess student-produced class diagrams based on an answer key. However, teachers have a problem with a lack of consistency in assessment as teachers can use different standards between answers. This research attempts to approach class diagram assessment automatically. The proposed approach consists of two assessments: semantic and structural similarities. Semantic similarity is calculated using lexical information in the class diagram, and structural similarity is calculated using the diagram's structure, ignoring its lexical information. Our results show that experts see semantic and structural similarities equally during assessment. The proposed approach shows substantial agreement with experts in class diagram similarity assessment. Therefore, the proposed approach can automatically assess class diagram similarity as reliably as experts can.