Structure-Aware Enhanced LLMs via Knowledge Graphs for Microservice Architecture Documentation

• Item type: SIG Technical Reports(1)
• 公開日: 2026-03-02
• タイトル / 言語: ja
• タイトル: Structure-Aware Enhanced LLMs via Knowledge Graphs for Microservice Architecture Documentation
• タイトル / 言語: en
• タイトル: Structure-Aware Enhanced LLMs via Knowledge Graphs for Microservice Architecture Documentation
• 言語: eng
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: School of Computing, Institute of Science Tokyo
• 著者所属: School of Computing, Institute of Science Tokyo
• 著者所属: School of Computing, Institute of Science Tokyo
• 著者所属(英): en
• 著者所属(英): School of Computing, Institute of Science Tokyo
• 著者所属(英): en
• 著者所属(英): School of Computing, Institute of Science Tokyo
• 著者所属(英): en
• 著者所属(英): School of Computing, Institute of Science Tokyo
• 著者名: Qiao,Lin; Profir-petru,Pârţachi; Takashi,Kobayashi
• 著者名(英): Qiao Lin; Profir-petru Pârţachi; Takashi Kobayashi
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Microservice documentation is critical yet hard to maintain due to its complex and distributed architecture. Developers use Large Language Models (LLMs) with Retrieval Augmented Generation (RAG) to automate the creation and update of microservice documentations; however, this approach has limitations. Standard RAG is blind to the high level, global structural dependencies among services and relies solely on semantic similarity (Structural Blindness). As a result, LLMs suffer from missing key dependencies and information during generation and leading to inaccurate documentation. We introduce a Graph RAG framework that integrates structural awareness into the generation process. A Knowledge Graph (KG) and a GNN are used to help retrieve the most relevant dependencies for the LLM during generation. This ensures the generated documentation is both semantically coherent and architecturally complete, addressing some of the current limitation. As the result, we improve on three metrics over our baselines: Correctness (0.66-1.64), Completeness (0.86-2.24), and Faithfulness (0.64-1.62).
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Microservice documentation is critical yet hard to maintain due to its complex and distributed architecture. Developers use Large Language Models (LLMs) with Retrieval Augmented Generation (RAG) to automate the creation and update of microservice documentations; however, this approach has limitations. Standard RAG is blind to the high level, global structural dependencies among services and relies solely on semantic similarity (Structural Blindness). As a result, LLMs suffer from missing key dependencies and information during generation and leading to inaccurate documentation. We introduce a Graph RAG framework that integrates structural awareness into the generation process. A Knowledge Graph (KG) and a GNN are used to help retrieve the most relevant dependencies for the LLM during generation. This ensures the generated documentation is both semantically coherent and architecturally complete, addressing some of the current limitation. As the result, we improve on three metrics over our baselines: Correctness (0.66-1.64), Completeness (0.86-2.24), and Faithfulness (0.64-1.62).
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN10112981
• 書誌情報: 研究報告ソフトウェア工学（SE）; 巻 2026-SE-222, 号 20, p. 1-8, 発行日 2026-03-02
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8825
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会