<![CDATA[Automated evaluation of flowchart representations is essential for the facilitation of the acquisition of basic programming concepts. Nevertheless, traditional evaluation systems that rely exclusively on structural matching demonstrate some of their most fundamental limitations. The false negative misclassification rates of such systems are frequently high when students create visually distinct structures for algorithmic logic that are semantically equivalent. A hybrid assessment framework is introduced in this study to improve the reliability and efficacy of code evaluation in order to address this challenge. The model that has been proposed combines the probabilistic feature extraction capabilities of Graph Convolutional Networks (GCNs) with mathematical logic verification through symbolic execution of an SMT Solver. While the SMT Solver deterministically establishes functional equivalence, the GCN module adaptively manages graph topological variations. Use of a real-world dataset consisting of 3.600 flowcharts generated by novice students was implemented to assess the hybrid system's functionality. According to quantitative experimental results, the proposed framework obtained a peak F1 Score of 0.88, which is a substantial improvement over conventional Abstract Syntax Tree (AST) methods (F1 Score 0.75). Additionally, the 77.4% reduction in false negative rates was achieved by incorporating the SMT Solver in comparison to a pure GCN configuration. Finally, the semantic equivalence and structural divergence issues that arise during algorithm assessment are effectively resolved by this dual architectural integration. By implementing the proposed system, higher education institutions are equipped with a more dependable mechanism for reducing human error, thereby improving the impartiality, accuracy, and efficiency of the evaluation process.]]>
Copyrights © 2026
