Requirements Engineering

Requirements Engineering (RE) is the systematic discipline concerned with the elicitation, specification, validation and management of the goals, functions and constraints of socio-technical systems. As one of the earliest activities in system development, it establishes a precise understanding of stakeholder needs, the environment in which a system will operate and the criteria by which success will be judged. Core RE activities include gathering and negotiating stakeholder expectations, modelling requirements in natural or formal notations, assuring quality through verification and validation, and maintaining traceability among requirements, design decisions and delivered artefacts. Over the past decade, increasing system complexity, ubiquitous connectivity and the advent of agile and machine-learning techniques have driven advances in automated requirement analysis, dynamic adaptation and socio-technical collaboration. RE now underpins safety-critical engineering, model-based development, software ecosystems and service-oriented architectures, ensuring coherent integration of heterogeneous components and continual alignment with evolving user, organisational and regulatory demands.

Research from Nature Portfolio

A novel deep-learning framework has been introduced for the automatic classification of non-functional requirements (NFRs), addressing the considerable effort involved in manual feature engineering. The proposed architecture, termed DReqANN, employs a multi-layer neural network that captures hierarchical feature structures and broader contextual information than traditional shallow models. Evaluated on two widely used NFR datasets comprising 914 instances, DReqANN achieved precision between 81 and 99.8 per cent, recall between 74 and 89 per cent and F1-scores between 83 and 89 per cent across various NFR categories. These results highlight the potential of deeper architectures to streamline NFR analysis and reduce analysts’ workload in large-scale RE deployments.

Research from all publishers

Zero-shot learning frameworks have been explored to overcome the scarcity of labelled data in requirement classification. By leveraging transformer-based language models and contextual embeddings, these approaches classify functional, non-functional and security requirements without task-specific training, achieving competitive F1-scores and demonstrating the feasibility of ultra-low-effort classification.

Domain-specific fine-tuning of pretrained transformers has yielded bespoke classifiers for aerospace requirements. A dedicated aerospace corpus enabled the development of aeroBERT-Classifier, which distinguishes design, functional and performance requirements with higher accuracy than generic models, despite limited annotated data. This work underlines the value of small, curated domain corpora.

Complementary research has harnessed natural language processing to automate the generation of Systems Modelling Language (SysML) diagrams from unstructured text. By extracting entities and relationships via rule-based and statistical methods, these pipelines produce initial structure and requirement diagrams automatically, offering engineers a standardised starting point for subsequent refinement.

Requirements Engineering publication trend

The graph below shows the total number of articles in requirements engineering across all publications each year (not limited to Nature Index journals).

Technical terms

Requirements Engineering (RE): The discipline of eliciting, specifying, validating and managing the goals, functions and constraints of a system in its operational environment.

Functional Requirement: A statement of a service or function that a system must perform, expressed in terms of behaviour.

Non-Functional Requirement: A constraint on system services or functions, often relating to quality attributes such as performance, security or usability.

Natural Language Processing (NLP): Computational techniques for analysing and extracting structured information from human language text.

Transformer: A deep-learning architecture using self-attention mechanisms to capture contextual relationships in sequential data.

Zero-Shot Learning: An approach that applies models to identify categories not seen during training by leveraging semantic descriptions or embeddings.

Systems Modelling Language (SysML): A general-purpose modelling language for systems engineering that supports specification, analysis, design and verification of complex systems.

References

1. Requirements Engineering.
2. A deep learning framework for non-functional requirement classification. Scientific Reports (2024).
3. Zero-shot learning for requirements classification: An exploratory study. Information and Software Technology (2023).
4. aeroBERT-Classifier: Classification of Aerospace Requirements Using BERT. Aerospace (2023).
5. Natural Language Processing for systems engineering: Automatic generation of Systems Modelling Language diagrams. Knowledge-Based Systems (2023).