Scholarlink

<p>With the increasing complexity of software systems, especially in large-scale agile systems, the need to efficiently bridge high-level requirements to architectural models has become significant. This study introduces a semi-automated, AI-based framework that converts user stories into architectural blueprints. It uses the framework of Natural Language Processing (NLP), clustering, and quality assessment to analyze user stories in a systematic way and identify major aspects of architecture and their relationships. A dataset of 245 annotated user stories based on healthcare, finance, and telecommunications areas, compiled by Mendeley, is used to calculate the scores of semantic relevance using Term Frequency-Inverse Document Frequency (TF-IDF) and Latent Dirichlet Allocation (LDA). The evaluation is performed using three baseline conditions, which include manual expert mapping, rule-based mapping without clustering, and TF-IDF-only mapping. The experimental outcomes show favorable improvements in comparison with all baselines: mapping accuracy rises to 89% compared to 62% (manual baseline), transformation time is reduced by 60%, and requirement quality and consistency are significantly improved. Compared to recent Large Language Model (LLM)-based pipelines, the suggested pipeline has a lower level of traceability and explainability, and it is computationally light. The results demonstrate the effectiveness of AI-based methods in the context of agile processes to enhance design traceability, minimize architectural drift, and derive intelligent architecture on a scale in the context of contemporary software development.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

From user stories to architectural blueprints: an AI-augmented NLP pipeline for large-scale agile software engineering

  • 期刊论文

  • 发表日期：

    2026年7月23日

查看全文
  • Amber Sarwar,
  • Huma Rauf,
  • Shariq Hussain,
  • Hala Mostafa,
  • Kashif Sultan,
  • Ibrahima Kalil Toure

摘要

With the increasing complexity of software systems, especially in large-scale agile systems, the need to efficiently bridge high-level requirements to architectural models has become significant. This study introduces a semi-automated, AI-based framework that converts user stories into architectural blueprints. It uses the framework of Natural Language Processing (NLP), clustering, and quality assessment to analyze user stories in a systematic way and identify major aspects of architecture and their relationships. A dataset of 245 annotated user stories based on healthcare, finance, and telecommunications areas, compiled by Mendeley, is used to calculate the scores of semantic relevance using Term Frequency-Inverse Document Frequency (TF-IDF) and Latent Dirichlet Allocation (LDA). The evaluation is performed using three baseline conditions, which include manual expert mapping, rule-based mapping without clustering, and TF-IDF-only mapping. The experimental outcomes show favorable improvements in comparison with all baselines: mapping accuracy rises to 89% compared to 62% (manual baseline), transformation time is reduced by 60%, and requirement quality and consistency are significantly improved. Compared to recent Large Language Model (LLM)-based pipelines, the suggested pipeline has a lower level of traceability and explainability, and it is computationally light. The results demonstrate the effectiveness of AI-based methods in the context of agile processes to enhance design traceability, minimize architectural drift, and derive intelligent architecture on a scale in the context of contemporary software development.