Scholarlink

<p>Non-functional requirements (NFRs) are vital in the development of healthcare systems where trust, compliance, and operational efficiency are paramount. However, NFR management is challenging due to their abstract nature, contextual variability, and diverse stakeholder interpretations. This research aims to develop a knowledge-driven semantic framework that facilitates systematic NFR management in healthcare systems by integrating semantic web technologies and AI-based reasoning to represent and manage NFRs. Its novelty lies in a two-layer ontology structure (healthcare domain and NFR-specific) combined with BERT-based classification and recommendation for semi-automated NFR elicitation, where the model suggests potentially missing NFR categories based on historical patterns. The framework is further distinguished by an empirical validation methodology combining ontology-based reasoning, ML-assisted elicitation, and stakeholder-driven evaluation in a real-world ICDSS case study. Dynamic conflict resolution is enabled through ML-based ranking that adapts mitigation priorities based on ontology context and previously accepted resolutions. It includes modules for NFR elicitation using machine learning techniques, ontology-based formalization for traceability and inference, and semantic integration for enhanced interoperability. A case study of an AI-driven remote patient monitoring system (ICDSS) for chronic heart failure was conducted to evaluate the framework. Compared to a baseline (manual NFR handling using traditional RE tools like spreadsheets and UML diagrams), evaluation results demonstrated a 64% increase in NFR coverage, identifying 9 additional distinct NFRs including explainability and compliance requirements, a 64% improvement in traceability score (from 2.8 to 4.6 on a 1–5 Likert scale), a 40% reduction in validation time (from 15 to 9&#xa0;h per iteration), and a 91% stakeholder satisfaction rate, where 91% of participants rated the framework ≥ 4 on a 1–5 Likert scale (measured via post-evaluation questionnaires on usability and effectiveness). The BERT classifier achieved an average F1-score of 0.87, outperforming SVM. By formalizing NFRs through ontologies and enhancing reasoning with AI, the framework contributes to more transparent, accountable, and trustworthy intelligent healthcare systems by improving measurable properties such as NFR traceability (from 2.8 to 4.6), conflict detection (from 21.4 to 30.4%), and stakeholder-validated clarity and auditability. It provides a scalable solution for managing NFRs and supports future extensions into real-time adaptive systems and broader clinical domains.</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

A knowledge-driven semantic framework for non-functional requirement engineering in intelligent healthcare systems

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月14日

查看全文
  • Anirban Bhar,
  • Sabnam Sengupta,
  • Sankhayan Choudhury

摘要

Non-functional requirements (NFRs) are vital in the development of healthcare systems where trust, compliance, and operational efficiency are paramount. However, NFR management is challenging due to their abstract nature, contextual variability, and diverse stakeholder interpretations. This research aims to develop a knowledge-driven semantic framework that facilitates systematic NFR management in healthcare systems by integrating semantic web technologies and AI-based reasoning to represent and manage NFRs. Its novelty lies in a two-layer ontology structure (healthcare domain and NFR-specific) combined with BERT-based classification and recommendation for semi-automated NFR elicitation, where the model suggests potentially missing NFR categories based on historical patterns. The framework is further distinguished by an empirical validation methodology combining ontology-based reasoning, ML-assisted elicitation, and stakeholder-driven evaluation in a real-world ICDSS case study. Dynamic conflict resolution is enabled through ML-based ranking that adapts mitigation priorities based on ontology context and previously accepted resolutions. It includes modules for NFR elicitation using machine learning techniques, ontology-based formalization for traceability and inference, and semantic integration for enhanced interoperability. A case study of an AI-driven remote patient monitoring system (ICDSS) for chronic heart failure was conducted to evaluate the framework. Compared to a baseline (manual NFR handling using traditional RE tools like spreadsheets and UML diagrams), evaluation results demonstrated a 64% increase in NFR coverage, identifying 9 additional distinct NFRs including explainability and compliance requirements, a 64% improvement in traceability score (from 2.8 to 4.6 on a 1–5 Likert scale), a 40% reduction in validation time (from 15 to 9 h per iteration), and a 91% stakeholder satisfaction rate, where 91% of participants rated the framework ≥ 4 on a 1–5 Likert scale (measured via post-evaluation questionnaires on usability and effectiveness). The BERT classifier achieved an average F1-score of 0.87, outperforming SVM. By formalizing NFRs through ontologies and enhancing reasoning with AI, the framework contributes to more transparent, accountable, and trustworthy intelligent healthcare systems by improving measurable properties such as NFR traceability (from 2.8 to 4.6), conflict detection (from 21.4 to 30.4%), and stakeholder-validated clarity and auditability. It provides a scalable solution for managing NFRs and supports future extensions into real-time adaptive systems and broader clinical domains.