AI-Driven Microservice Identification for Business Process Digital Transformation: A Multi-Dependency Collaborative Clustering Approach
Mohamed DAOUD, Fatima ezzahra Assamid, Assia Ennouni, My Abdelouahed Sabri · Statistics Optimization & Information Computing · 2025
In the era of digital transformation, enterprises are increasingly seeking to modernize their legacy monolithic systems in favor of more agile and modular architectures. Microservices have emerged as a compelling solution, offering scalability, maintainability, and independent deployment. However, the automated extraction of microservices from legacy systems—particularly when documentation is sparse or outdated—remains a complex and unresolved challenge. This paper introduces an AI-powered, multi-view methodology for microservice identification based on business process (BP) models. Unlike traditional approaches that rely on static code analysis or single-view aggregation, our method simultaneously captures and analyzes three critical types of dependencies within business processes: control flow, data exchange, and semantic similarity. Each dependency is modeled separately and processed through a collaborative clustering framework, where AI agents exchange signals to achieve consensus-based service decomposition. Artificial Intelligence plays a dual role in our system: it is used for semantic enrichment—via Natural Language Processing (NLP) and Sentence-BERT embeddings—and for optimizing the clustering strategy through dependency alignment and explainability metrics. A real-world case study on the Bicing bike-sharing system in Barcelona, composed of over 50 business activities, demonstrates the effectiveness and scalability of our approach. Experimental results show that the AI-enhanced model achieves superior clustering performance in terms of cohesion, coupling, and interpretability compared to baseline methods. By integrating AI-driven analytics with business process understanding, our approach provides a robust pathway toward automated, explainable, and domain-aligned microservice extraction—supporting sustainable digital transformation at scale. Une étude de cas réelle sur le système de partage de vélos Bicing à Barcelone, composé de plus de 50 activités commerciales, démontre l'efficacité et l'évolutivité de notre approche. Les résultats expérimentaux montrent que le modèle amélioré par l'IA atteint des performances de regroupement supérieures en termes de cohésion, de couplage et d'interprétabilité par rapport aux méthodes de base.