ASTutor: An Architectural Design for Automated and Intelligent Evaluation of Programming Tasks

Icochea N.M.G.; Ochoa C.A.C.; Silva P.D.R. · Proceedings - 2025 5th International Conference on Information Systems and Software Technologies, ICI2ST 2025 (2025) · DOI: 10.1109/ici2st68465.2025.00011

Resumen

In the university context, the evaluation of programming exercises continues to be a manual, repetitive task prone to inconsistencies. Although solutions exist that address specific aspects such as functional testing or plagiarism detection, there is currently no comprehensive application oriented to the educational context that automates grading considering multiple criteria configurable by the instructor. This problem directly impacts instructors, who must individually review dozens of submissions per task, evaluating aspects such as syntax, semantic structures, and code originality. This burden limits the possibility of offering timely and objective feedback to students. This article presents the architectural design of an intelligent web application for automated evaluation of programming exercises in Java. The proposal integrates three main modules: plagiarism detection, semantic structure detection, and compilation validation. Each module relies on specialized techniques, such as Abstract Syntax Trees (AST), recursive encoders, and siamese neural network architectures. The grading logic will be configurable by the instructor, combining the results of each module into a weighted evaluation. The main contribution of this article is the detailed definition of the functional architecture of the application and the evaluation logic, supported by preliminary individual validation of the proposed modules.

Instituciones: Universidad Peruana de Ciencias Aplicadas