Monolith Refactoring via Attributed Graph Embeddings

The patent replaces manual mechanical analysis with automated machine learning-based graph embedding systems. The system automatically generates graph representations of the monolith application, computes node embeddings using GCN or GAT models, and performs clustering without human intervention, thereby substituting time-consuming manual efforts with automated computational processes while maintaining or improving accuracy.

The system enables self-service refactoring analysis by automatically performing all necessary steps: generating application graphs from source code, computing embeddings, identifying domain model boundaries, and suggesting microservice partitions. Developers can initiate the process and receive results without requiring deep expertise in refactoring methodologies, making the system serve itself and the developers efficiently.

The patent segments the complex refactoring process into distinct automated modules: (1) source code parsing and graph generation, (2) node embedding computation using pre-trained GCN/GAT models, (3) constraint identification, (4) clustering execution, and (5) result visualization. Each module handles a specific aspect independently, reducing overall process complexity while maintaining high clustering accuracy through specialized processing at each stage.

The system introduces graph embeddings as an intermediary representation between raw source code and final clustering decisions. Instead of directly analyzing complex source code structures, the system transforms code into graph representations, computes meaningful embeddings that capture structural and behavioral features, and then performs clustering on these embeddings. This intermediary step simplifies the overall process while preserving critical information for accurate clustering.

The system performs preliminary identification and encoding of constraints before executing the clustering algorithm. During graph generation and embedding computation, the system pre-identifies important application invariants such as database ownership, external service dependencies, and transaction boundaries, and encodes them as soft or hard constraints. This preliminary action ensures that constraints are considered during clustering, maintaining reliability while allowing automated fast execution.

The system implements feedback mechanisms where clustering results are evaluated against identified constraints, and the process can be iteratively refined. The constraint satisfaction status is fed back into the clustering optimization, allowing the system to adjust cluster assignments to better respect application invariants. This feedback loop maintains high productivity through automation while ensuring reliability through constraint adherence.