Recognition of roles of variables based on deep learning technologies

Using variables in regulating computing procedures is a key competence of structured programming. This tacit knowledge, which is highly repetitive and transferrable between programming tasks, may be succinctly described as the roles of variables. The role of a variable defines the pattern of the uses and changes of variable for computing procedures. This notion is an effective pedagogical tool for providing feedback on how to fix faulty computing procedures linked to misuse of variables. This paper proposes the automation of variable role recognition and describes an evaluation study for a variable role classifier based on deep neural network architectures. The classifier is designed to learn the variable roles from execution-time traces of variables. Experimental results showed the classification accuracy reached over 0.95. Other findings indicated that the variable roles revealed diverse temporal characteristics in their execution-time traces, and specifically designed deep learning architectures would be needed for some individual roles. Automated variable role recognition will find applications in intelligent tutoring systems and the feedback to programs will significantly improve learning effectiveness.