Identifying Evidences of Computer Programming Skills Through Automatic Source Code Evaluation

This thesis is contextualized in the teaching of computer programming in Computing courses and investigates aspects and strategies for automatic and continuous evaluation of student developed source codes. The state of the art was identified through systematic literature review and revealed previous research tends to perform evaluations based on source codes technical aspects, such as functional correctness assessment and error detection. Skills-based assessments, in turn, are less explored although having potential to provide details of skills represented by high-level concepts, such as conditionals and repetition structures. A method for automatic identification of learning evidences is then proposed as a skills-based approach to automatic evaluation of programming source codes. The method is characterized by implementing different strategies for source code evaluation, identifying evidences of programming skills, and representing these skills in a student model. Experiments conducted in controlled scenarios (testing datasets) have shown automatic source code evaluation strategies are viable. Experiments conducted in real scenarios (student-made source codes) produced results similar to controlled scenarios, however, implementation-related limitations were revealed for some strategies, such as vulnerabilities to unexpected syntax and flaws in regular expressions. A skill set was selected to compose our student model, represented by a Dynamic Bayesian Network. Experiments have shown feeding the model with evidences resulting from source codes automatic evaluation allows monitoring students’ skills progress. Finally, automatic strategies coupled with student model capabilities enabled demonstrating skills-based assessment, which showed a valuable resource for identifying functionally correct source codes, but conceptually incorrect; when a program is correct functionally, returning expected results to specific inputs, but it was built with erroneous concepts and resources.