Semantic Approaches to Microservice Identification: A Systematic Literature Review

Abstract

Decomposition of monolithic systems into microservice architectures has become a critical challenge in modern software engineering. Whereas traditional approaches rely primarily on structural analysis, emerging semantic approaches promise more accurate decomposition by incorporating the meaning-aware analysis of software artifacts. This systematic literature review presents the first comprehensive analysis of semantic approaches for microservice identification, examination techniques, evaluation methodologies, and practical adoption challenges. The Scopus database was systematically searched for semantically specific terms. After applying strict inclusion criteria, 17 high-quality papers, representing a 48.6% selection rate. Each study was analyzed for its technical approach, evaluation methodology, and practical applicability. Four distinct categories were identified: NLP-based (41%), hybrid (35%), ontology-based (18%), and AI-enhanced (6%). Hybrid approaches demonstrated superior performance (F1-score: 0.81), whereas domain specific models showed 15-20% improvement in accuracy. The selected studies were published between 2020-2025, with the majority concentrated in 2025. This temporal concentration reflects the emergence of semantic approaches as distinct research areas. Our search covered 2010-2025 but found limited genuine semantic work in earlier years, indicating that semantic microservice identification is a recent phenomenon as evidenced by the current research surge. Semantic approaches consistently outperform traditional methods, achieving 23-45% improvements in coupling reduction and 15-38% enhancements in cohesion. However, challenges remain, including the lack of standardized evaluation benchmarks, limited tool maturity, and insufficient industrial validation. The field must focus on benchmark development, production-ready tools, and industrial case studies.