<![CDATA[This research aims to reconstruct the microservices architecture of the Moodle monolithic system using the Graph-Based Deep Learning approach, especially the Graph Convolutional Network (GCN). The study focused on analyzing dependencies in 13,841 PHP files from the Moodle repository version MOODLE_403_STABLE. The main problem is the high complexity of dependencies in Moodle's monolithic architecture, which makes it difficult for traditional clustering methods to automatically determine the right service boundaries. The results showed that the Wide GCN configuration (with a wider layer) converted the negative modularity (-0.29007) to positive (0.11676) in the global population and achieved the highest modularity value of 0.24794 in the assessment module. The study's main finding is that the use of Wide GCN consistently improves cohesion by up to tenfold compared to monolithic conditions, although there is a trade-off in the form of increased coupling. The contributions of this study include: (1) empirical evidence regarding the advantages of GNN in systems with dense dependencies, (2) systematic comparisons between wide and standard configurations, and (3) simultaneous integration of four architecture quality metrics.]]>
Copyrights © 2026
