Ana gezinime geç Aramaya geç Ana içeriğe geç

An Empirical Analysis of LLM-Driven Refactoring for Microservices

  • Istanbul Technical University
  • University of Southern Denmark

Araştırma çıktısı: Kitap/Rapor/Konferans Bildirisinde Bölüm › Konferans katkısı › Hakemli

Özet

Large Language Models (LLMs) are increasingly used in software engineering, raising the question of whether they can reliably support refactoring in microservice-based systems, where architectural constraints and deployment requirements make refactoring particularly complex. This study empirically examines LLM-driven refactoring using a structured, three-phase prompting protocol applied twice to four microservices previously labeled as low-maintainability by experienced developers. Three state-of-the-art LLMs, GPT-5.2 Extended Thinking, Claude Opus 4.5, and Gemini 3 Pro, were evaluated across analysis, refactoring, and self-audit phases, followed by deployment, execution, and metric-based validation. Of 24 total refactorings, 22 successfully passed build and run verification, showing that LLM-supported refactoring is feasible in real microservice environments. Across all models, refactored microservices demonstrated consistent improvements in software quality. Maximum cognitive complexity decreased by about 45%, maximum cyclomatic complexity by around 31%, and average cognitive complexity by 26%, while code duplication decreased by roughly 9%. Expectedly, LOC and NoM increased due to the enforced method-decomposition strategy. Model behaviors varied: Claude Opus 4.5 produced the strongest overall reductions in complexity, and Gemini 3 Pro achieved competitive improvements with the smallest code growth. Overall, the findings indicate that LLM-driven, developer-assisted refactoring can produce measurable and functionally safe improvements in microservice-based systems.

Orijinal dilİngilizce
Ana bilgisayar yayını başlığıProceedings - 2026 IEEE 23rd International Conference on Software Architecture Companion, ICSA-C 2026
YayınlayanInstitute of Electrical and Electronics Engineers Inc.
Sayfalar369-376
Sayfa sayısı8
ISBN (Elektronik)9798331557409
DOI'lar
Yayın durumuYayınlandı - 2026
Etkinlik23rd IEEE International Conference on Software Architecture Companion, ICSA-C 2026 - Amsterdam, Netherlands
Süre: 22 Haz 2026 → 26 Haz 2026

Yayın serisi

AdıProceedings - 2026 IEEE 23rd International Conference on Software Architecture Companion, ICSA-C 2026

???event.eventtypes.event.conference???

???event.eventtypes.event.conference???23rd IEEE International Conference on Software Architecture Companion, ICSA-C 2026
Ülke/BölgeNetherlands
ŞehirAmsterdam
Periyot22/06/26 → 26/06/26

Bibliyografik not

Publisher Copyright:
© 2026 IEEE.

Parmak izi

An Empirical Analysis of LLM-Driven Refactoring for Microservices' araştırma başlıklarına git. Birlikte benzersiz bir parmak izi oluştururlar.

Alıntı Yap