A Multi-View Approach for Automatic Microservice Extraction Based on BERT Models and Self-Weighted Clustering

Mogollon S.; Sarmiento E. · Proceedings - International Conference of the Chilean Computer Science Society, SCCC (2025) · DOI: 10.1109/sccc67219.2025.11420630

Resumen

The evolution from monolithic architectures to microservices remains an open challenge, particularly when functional and structural boundaries are not clearly defined. Although several automated approaches have been proposed to address the migration process, most rely on a single system view - such as models or use cases, source code, or API specifications - thus limiting their ability to capture the functional and semantic complexity of the domain. This paper presents an automated multi-view approach for microservice extraction, named MIDAS, which leverages BERT-based semantic representations and a self-weighted clustering scheme. MIDAS integrates three complementary system views - structural, semantic, and functional - into a unified similarity graph that improves the identification of microservice boundaries. Initial evaluations on the JPetStore and TFWA benchmark systems demonstrate that MIDAS achieves competitive performance in terms of structural modularity, coupling, and functional cohesion, outperforming representative state-of-the-art approaches. The implementation is available at https://github.com/i-am-sergio/midas-m2m.

Instituciones: Universidad Nacional de San Agustín de Arequipa