Developing the skills required for requirements engineering students to conduct effective requirements elicitation interviews is critical yet challenging, as it requires the development of soft skills in addition to technical knowledge. Role-playing is widely adopted in requirements engineering education to support the development of these skills but is often constrained by time and resource limitations. Although recent advances in large language models (LLMs) enable automated and interactive stakeholder simulations for role-playing, their application in requirements engineering education remains limited by hallucinations and inconsistent responses. To address these limitations, this study investigates the use of retrieval-augmented generation (RAG), implemented using the LangChain framework, to support requirements elicitation activities in a requirements engineering course. We conducted controlled experiments with 69 students, comparing cohorts who interacted with non-technical faculty role-players and those who engaged with RAG-based simulated stakeholders. The results indicate that while students perceived invited stakeholders as more realistic and engaging, the RAG-based automated stakeholder produced higher-quality elicitation questions and more complete feature identification.
Requirements engineering is crucial for software development. Analyzing large sets of requirements in natural language is difficult and time-consuming, especially in Internet of Things (IoT) projects, where heterogeneous devices, services, communication infrastructures, and environmental constraints increase the complexity of the requirements. In this context, clustering can support requirements analysis by organizing functional requirements into semantically coherent groups that represent the same functional aspect or subsystem. This helps analysts reduce manual effort, identify relationships and dependencies between requirements, structure large sets of requirements, and better understand system responsibilities and specific IoT concerns. For these reasons, we propose ReqCluster4IoT, a method for clustering functional requirements in IoT applications. The method builds on a focused exploration of the literature to identify IoT characteristics, along with experiments on requirement classification and text semantic similarity. We evaluated the method’s acceptance and usefulness through an online questionnaire and a focus group. The online questionnaire collected feedback from 10 IoT software developers, whereas the focus group provided more in-depth insights into the method’s results. Findings suggest that ReqCluster4IoT can capture meaningful relationships among requirements and generate clusters that practitioners perceive as useful. Additionally, the method helps practitioners identify how related functionalities are organized into clusters, reveal relationships among requirements, and associate each cluster with IoT-specific characteristics, thereby supporting requirements analysis, implementation planning, and requirements management. Overall, participants reported positive acceptance of the method regarding its usefulness, its potential to reduce effort, and their intention to use it.
Software is a crucial component of modern everyday life. However, ethical issues in software, such as privacy issues, censorship and behavior manipulation are prevalent in software applications. To develop ethical software, it is essential to understand the ethical concerns of its end-users. To provide these insights, we conducted a survey among 725 participants and analyze which ethical concerns they have regarding software and why, as well as the actions they are willing to take when confronted with ethical issues in software. The results indicate that privacy, scam and misinformation are important ethical concerns for the wide majority, with scam prompting the strongest reaction in all unethical software scenarios. Reasons for these concerns included—but are not limited to—the exposure of personal sensitive information, perceived societal harm and fear of material loss. The most common reactions to ethical issues in software are to stop using the software and encourage friends and family to do the same. We also find that perceived importance and reactions to unethical scenarios in software products varies significantly by gender, education, and continent.
This research commentary explores how requirements engineering (RE) can contribute to achieving digital humanism in AI-enabled healthcare, ensuring that technological innovation is balanced with human values. Drawing on insights from a panel discussion at the REWBAH’25 (RE for Well-Being, Aging, and Health) workshop, we identify four key themes: the necessity of interdisciplinary collaboration, the nuances of cultural and individual differences, human-AI collaborative decision-making, and the challenge of balancing technological advancement with sustainability and emotional considerations. We argue that RE is well suited for addressing these issues because it brings together diverse stakeholder perspectives to define what systems should achieve and how they should be used. In this paper, we build on digital humanism principles, on RE-relevant frameworks for AI, and on the above themes to frame the role of RE in supporting human-centred healthcare. We highlight open challenges, and outline five promising research directions and related research questions to guide future research and practice in RE, aiming to ensure that AI-enabled healthcare systems better reflect the values of digital humanism.
Requirements Engineering (RE) is a socio-technical process that relies on effective communication among stakeholders. Understanding how national cultural influences affect RE activities is crucial, given participants’ diverse cultural backgrounds. This paper presents a refined framework for identifying cultural influences (CIs) on RE-related activities and a web-based application developed to support it. The refined framework incorporates the latest Hofstede scores and covers nearly double the CIs compared to its earlier version. We evaluated the proposed framework against in-depth case studies from 8 countries (Australia, Thailand, China, Pakistan, South Korea, Saudi Arabia, Vietnam, and Bangladesh), where the CIs have been identified on the basis of surveys and interviews. Analysis showed the proposed framework consistently provided better coverage than the original. We extended the framework by assigning evidence-based confidence levels to each CIs based on the evaluation wrt. primary case studies. Our approach allows RE practitioners to evaluate and manage cultural impacts effectively. The framework provides a comprehensive analysis of national cultural influences and delivers more accurate and contextually relevant insights. The tool aids users in obtaining relevant national CIs, confidence levels, and summaries for a selected country. This enables stakeholders, practitioners, researchers, and educators to manage cultural dynamics effectively, enhancing RE-related activities.
Estimating the effort required for software development is challenging due to the inherent complexity of software. The use of Large Language Models (LLMs) could facilitate this process for development teams. This study aims to explore the capability of LLMs to estimate the effort needed to complete requirements expressed as user stories in software development. The estimates of 10 LLMs and 42 developers were evaluated using the agile T-shirt Sizing technique for 12 user stories. The preservation of the expected logical order, less effort with greater experience and optimism were analyzed using statistical tests. Subsequently, absolute accuracy was compared with the actual effort obtained from post-mortem developments of a small e-commerce business. Human developers and several LLMs showed similar accuracy in their estimates compared to the actual post-mortem effort. Large models, those with hundreds of billions of parameters, consistently preserved the expected order in estimates, reducing estimated effort by increasing experience and optimism. The experience level showed a greater influence than the degree of optimism. Some smaller models, commonly referred to as Small Language Models (SLMs) ranging from four and eight billion parameters, showed inconsistent patterns, so model size was more decisive than reasoning capability for the quality of estimates. These findings suggest that the effort estimates from some LLMs are more accurate than those from human developers when the developer’s experience level and degree of optimism are properly incorporated.
Defining and reasoning about goals in multi-actor dialectical systems requires formalizing often incomplete or unclear stakeholder requirements within a goal model. This process involves actor dialogue to clarify assumptions, resolve gaps, and justify modeling decisions, making argumentation central to validating decisions collaboratively. This work explores how argumentation theories can capture the informal reasoning of multiple actors by integrating two frameworks: Argumentation-based Proof-Events (APEC), which treats goal-oriented problem solving as a social and temporal process, and Provers’ System (PS), which models the internal reasoning and meta-level attributes of individual actors. By combining these, the proposed APEC-PS framework bridges internal cognitive reasoning with external collaborative dynamics, offering a more comprehensive view of actor interactions and argument structures. Rooted in symbolic AI, it employs formal logic and structured representations to enable goal-oriented defeasible reasoning. This integration is particularly well-suited for managing evolving knowledge and uncertainty in open-ended decision-making contexts. We utilize a domain-agnostic example from mathematical practices, exemplified by the Mini-Polymath 4 project, to showcase how the model’s abstract and generic nature can support diverse goal-oriented applications through dynamic actor interactions without being constrained by specific domain aspects.
This study empirically assesses how effectively two Large Language Models (LLMs), OpenAI o3-mini and DeepSeek R1, can transform terse feature-request titles from open-source software (OSS) issue trackers into well-formed software requirements. It further examines how prompt-engineering strategies shape requirement quality and evaluates a scalable “LLM-as-a-Judge” approach for automated quality assessment based on three ISO/IEC/IEEE 29148:2018 quality attributes: Unambiguity, Verifiability, and Singularity. We extract 150 feature-request titles from five OSS repositories and pair each title with every combination of two LLMs and three prompt styles, producing 900 candidate requirements. An independent evaluator LLM rates each requirement on these three quality attributes using a Likert scale and provides textual rationales. Ordinal data are analyzed with descriptive statistics and non-parametric tests, complemented by thematic analysis of the evaluator’s explanations. A targeted human validation study with five evaluators on a stratified sample of 50 requirements assesses the reliability of the automated judge. LLMs often produce high-quality requirements; however, scores vary with input clarity and prompt design. Few-shot prompting consistently boosts Singularity, while Expert Identity prompting sometimes raises Verifiability but frequently harms Singularity. o3-mini shows a modest but significant edge in Singularity. Qualitative review echoes these patterns, revealing trade-offs between added detail and focus. The LLM-as-a-Judge protocol delivers consistent, scalable evaluations whose scores correlate significantly with aggregate human judgment ( p < 0.05 for all three attributes). Modern LLMs can expedite the drafting of initial software requirements from informal OSS inputs, but their output quality hinges on careful prompt selection and the inherent clarity of the source title. Prompt effects are model-dependent and may introduce trade-offs among quality attributes, so human oversight remains indispensable for refinement. The LLM-as-a-Judge framework proves a practical, human-validated technique for large-scale evaluation, enabling rapid, reproducible insights into LLM-driven requirements engineering workflows.
Large Language Models (LLMs) have the potential to automate knowledge-intensive interactions in enterprise systems, yet their adoption is often limited. One reason is a lack of user trust. This study examines how trust can be systematically engineered into an LLM-driven, multi-agent chatbot that handles routine human-resources (HR) queries. We follow a two-cycle Design Science Research methodology. Cycle 1 triangulated a systematic literature review with a thematic analysis over semi-structured interviews of six employees at a global firm and a confirmatory workshop with five AI experts to elicit and validate trust requirements. Cycle II instantiated these requirements in a multi-agent LLM chatbot prototype artifact and evaluated whether the artifact satisfies them through controlled user sessions and expert walkthroughs, emphasizing perceived usefulness and trust captured in post-task interviews ( n = 11 ) and operationalizing trust via alignment-oriented measures (faithfulness, answer relevancy, and adversarial robustness). The study yields a refined taxonomy of external (transparency, organizational safeguards, third-party security) and internal (model provenance, bias risk, reliability) trust factors, identifying reliability as the primary determinant of adoption. The implemented design achieved ≥ 0.86 on trust-aligned metrics and was endorsed by 9/11 participants as ready for field deployment. These findings demonstrate that trust can be proactively addressed through design and offer prescriptive guidelines for software engineers seeking to embed LLMs safely and responsibly in socio-technical contexts.
Alliances of organizations (joint ventures (JV) and merger acquisition (M A)) have increasingly been formed to achieve strategic business goals and, as a consequence, large, complex, software-intensive Systems-of-Systems (SoS) have emerged from the complicated interactions among software systems of different organizations/companies. By presenting unique characteristics, in particular, the emergent behavior, evolutionary development, and dynamic reconfiguration, SoS present a big challenge for the elicitation of their requirements, mainly due to ever-changing requirements that continuously emerge at run-time. At the same time, important information contained in the business process level that could serve as a source of these requirements has been neglected. The main contribution of this work is to present a novel source of information, referred to as Processes-of-Business Processes (PoP), to collect requirements of SoS. PoP refer to dynamic, large, and complex business processes formed from the complicated interplay of business processes of organizations that take part in alliances. We also present RE-PoP, a method that systematically extracts SoS requirements from PoP. To evidence the benefits of PoP and RE-PoP, we applied them in a real-world case study involving the Brazilian public health system. By providing a holistic, accurate view of large business processes, PoP together with RE-PoP can assist requirement engineers in the complicated task of identifying tangled requirements of SoS and its constituent systems, as well as providing traceability among them. Information from the business process level should be systematically used to derive SoS requirements, aiming at promoting the SoS sustainability and longevity that are tightly impacted by the business process level.
Requirements traceability, the process of establishing and maintaining relationships between requirements and various software development artifacts, is paramount for ensuring system integrity and fulfilling requirements throughout the Software Development Life Cycle (SDLC). Traditional methods, including manual and information retrieval models, are labor-intensive, error-prone, and limited by low precision. Recently, Large Language Models (LLMs) have demonstrated potential for supporting software engineering tasks through advanced language comprehension. However, a substantial gap exists in the systematic design and evaluation of prompts tailored to extract accurate trace links. This paper introduces TraceLLM, a systematic framework for enhancing requirements traceability through prompt engineering and demonstration selection. Our approach incorporates rigorous dataset splitting, iterative prompt refinement, enrichment with contextual roles and domain knowledge, and evaluation across zero- and few-shot settings. We assess prompt generalization and robustness using eight state-of-the-art LLMs on four benchmark datasets representing diverse domains (aerospace, healthcare) and artifact types (requirements, design elements, test cases, regulations). TraceLLM achieves state-of-the-art F2 scores, outperforming traditional IR baselines, fine-tuned models, and prior LLM-based methods. We also explore the impact of demonstration selection strategies, identifying label-aware, diversity-based sampling as particularly effective. Overall, our findings highlight that traceability performance depends not only on model capacity but also critically on the quality of prompt engineering. In addition, the achieved performance suggests that TraceLLM can support semi-automated traceability workflows in which candidate links are reviewed and validated by human analysts.
Although the practices proposed in agile methods to speed up software development are beneficial to the organization, they can affect how requirements engineering activities are conducted as these practices are not explicit. These activities, when not carried out properly, can result in poorly understood, omitted, and poorly specified requirements, leading to requirements debt and can compromise other stages of software development. However, requirements debts, as well as practices to mitigate them, are little explored and, on many occasions, are unknown, not monitored, or managed efficiently, which causes high maintenance costs over time. The objective of this article is to present an Ontology focused on requirements debt in the agile context, enhancing practices that can mitigate them during agile requirements engineering activities. The research was conducted through an investigation of the state of practice supported by grounded theory, unifying the result with information obtained in two systematic studies of the literature: a systematic mapping of the literature that consists of identifying the causes, consequences, and practices focused on requirements debt and a tertiary study focused on debt management, supporting the knowledge obtained about requirements debt and serving as input for the construction of an ontology, called “OntoReD”. As a result of this study, an ontology named OntoReD was developed, aimed at representing and managing requirements debt in agile contexts. It centralizes knowledge about the causes, consequences, and mitigation practices, in addition to providing guidelines for their application in the software development process. OntoReD also establishes relationships among causes, consequences, and practices, enabling the identification of the origins of the debt and ways to prevent or reduce it. For the scientific community, OntoReD represents a relevant conceptual contribution, serving as a foundation for future studies and allowing adaptations to other types of technical debt. In the organizational context, it can be integrated into the development of a tool to support professionals in agile organizations in understanding the current scenario, offering practice recommendations through an interactive interface. This approach can support more informed and effective decisions in the management, prevention, and reduction of requirements debt.
Decentralized applications (DApps) are applications running on a blockchain and enable what has been coined Blockchain 3.0. They represent an evolution in digital technology, offering transparent, secure, and unmediated solutions. However, despite their considerable potential, the adoption of DApps remains limited. With this article, we aim to identify the main obstacles to DApps adoption, and propose recommendations for DApps developers, including recommendations about improving the User Experience (UX) and facilitating onboarding within these applications. Through qualitative interviews with experts, followed by the analysis of 25 DApps, we elicit and analyze the necessary requirements to address these issues. This paper has both practical and theoretical implications. Specifically, the resulting requirements will guide developers and entrepreneurs when designing a DApp by highlighting the various functional and non-functional requirements to take into account to increase the likelihood of user adoption. Furthermore, the findings contribute to the body of knowledge pertaining to the software engineering of DApps, a field that has been recognized as important by the research community. Notably, we discuss requirements engineering- and UX-related challenges, as well as best practices to implement for better DApp adoption.
The utmost importance of privacy and security requirements in software development calls for adopting methods that enable the identification and proactive mitigation of these issues during the system development. Our survey of 45 primary studies provides an overview of the methods, document types, and datasets employed in tackling this challenge, along with an analysis of approaches demonstrating superior performance based on document types and specific identification problems. Analysis reveals a wide adoption of ML-based systems on diverse datasets, showcasing the effectiveness of leveraging various sources of information to identify privacy and security requirements in software development.
In today’s rapidly evolving markets, the pressure to accelerate software delivery while ensuring alignment with stakeholder needs is paramount. While Agile methodologies and Low-Code platforms have expedited the delivery process, the quality of requirements specification remains a critical component of project success. Poorly defined requirements necessitate repeated clarification cycles for developers, leading to sprint delays and increased project costs. To address this challenge, we propose the Large Language Model-Based Agents for Requirements Specification (LLMBA4RS) method, developed following a design science research approach. This method leverages Retrieval Augmented Generation (RAG) and the CrewAI framework to assist requirements engineers in crafting user stories based on minimal functional requirements. Through the demonstration and evaluation of three applications built on Low-Code platforms, practitioners assessed the method’s effectiveness in generating consistent requirements and suggesting related user stories.
Software Engineering encompasses all aspects of producing useful and quality software. Agile practices, widely adopted for their principles of incremental delivery, continuous improvement, and close collaboration between development teams and business experts, face challenges in enhancing process quality and capability to meet customer demands and inspire trust. The Agile Quality Framework (AQF), which includes the Quality Agile Model (QuAM) and the QuAGI software tool, addresses these challenges by facilitating the measurement and monitoring of various aspects of agile development processes. This paper emphasizes the importance of requirements management within agile projects, particularly through Component No. 4 of QuAM, which focuses on defining needs as user stories. Effective requirements management is crucial in agile environments where dynamic customer participation and continuous product deliveries can introduce significant risks if not properly managed. Previous validation experiences using AQF have highlighted the adverse impacts of incorrect elicitation, analysis, specification, and validation of requirements. Despite the flexibility of agile practices, the full extent of their solutions to traditional Requirements Engineering (RE) problems remains unclear. There is a recognized dissociation between agile practices and RE practices, which, if integrated, could resolve common issues and improve the overall quality of both the process and the final product. This paper presents a systematic literature review to analyze and evaluate the applicability of traditional RE practices in agile software development (ASD) processes. The goal is to propose effective methods or practices for requirements management in agile contexts to enhance the quality of software development processes and products.
Context: Requirements classification is a fundamental task in software engineering that involves classifying natural language requirements into predefined classes, such as functional and non-functional requirements. Accurate classification is critical to the success of software projects, as it helps reduce development risks and improve overall software quality. However, most existing classification models rely on supervised learning approaches, which demand large, annotated datasets that are resource-intensive, time-consuming, and dependent on domain-specific expertise to acquire. Moreover, these models often struggle with generalizability, as they typically require retraining or fine-tuning for each new classification task. Objective: This study aims to address the challenge of dataset scarcity in requirements classification by investigating the effectiveness of prompt-based LLMs. Method: We conducted an empirical benchmarking study to evaluate and statistically compare the performance of multiple prompt-based LLMs and prompting techniques across several requirements classification tasks on two English requirements classification benchmark datasets: PROMISE and SecReq. The evaluation incorporated four prompting techniques: zero-shot, few-shot, persona, and chain of thought (CoT) prompts. For each task, we assessed and statistically compared the performance of different model–prompt configurations. We then compared the best-performing prompt-based LLM configurations on each task against the performance of the state-of-the-art fine-tuned transformer-based model. Results: The results demonstrate that prompt-based LLMs, particularly when paired with few-shot prompting, can achieve classification performance comparable to or even exceeding that of the state-of-the-art fine-tuned transformer-based model. Furthermore, augmenting few-shot prompts with either persona alone or a combination of persona and CoT can potentially improve model performance. Conclusion: This study highlights the potential of prompt-based LLMs as a practical and scalable alternative to traditional supervised approaches for requirements classification. By reducing the dependency on large annotated datasets, these models provide a flexible solution to overcoming data scarcity and enhancing the generalizability of automated requirements classification tools.
Use case modelling is an important technique used in software engineering and systems analysis to define, clarify, and organize system requirements from a user’s perspective. A Unified Modelling Language’s (UML) use case diagram is a graphical representation of the interactions between actors (users or external systems) and a system under consideration. One of the primary challenges in use case modelling is the extraction of essential elements for generating use case diagrams from natural language documents. This paper presents an automated approach by leveraging Natural Language Processing (NLP) techniques and rule based approach, to syntactically parse specifications and expedite the process. Specifically, the Subject-Verb-Object (SVO) relationship extractor is applied to establish relationships between subjects (actors) and verb + object (use case) pairs within a sentence boundary. The approach identifies relevant actors and use cases, the crucial components of a use case diagram, from the parsed data. The approach is implemented using the SmartNet dataset and validated against state-of-the-art case studies namely Philips and E-store, to evaluate its effectiveness and accuracy in extracting actors and use cases. The performance of the proposed approach on the SmartNet dataset yields an accuracy of up to 84
The failure of many software projects can be due to the poor quality of the software requirements. Furthermore, it is possible that these requirements are a consequence of, among other issues, poor identification of the stakeholders in the project which will undoubtedly lead to incorrect preliminary requirements. Therefore, the training of the practitioners who will carry out the activities related to the elicitation of requirements in real projects, such as the stakeholder identification, is crucial to efficiently determine a correct list of needs and expectations for the software that will be developed. In this regard, the landscape on Requirements Engineering education has changed over the last decade and serious games have been continuously used to improve crucial factors in the efficient practice of this discipline, such as motivation and communication skills. With this aim in mind, this study introduces the “Wonderland Mystery” game, which is based on the Volere Stakeholder Analysis template, a practical tool for helping practitioners in identifying the stakeholder roles and the types of knowledge that is required from them in a given project. An empirical evaluation involving 76 fifth-year Computer Engineering undergraduates was carried out in order to explore their perceptions of this learning alternative. Data were collected through questionnaires, interviews, and game sessions. The results obtained demonstrate the game’s capacity to strengthen the students’ relevant skills on stakeholder identification, while their confidence, satisfaction, and perceived learning are significantly improved considering information collected from pre and post-evaluations.
Software traceability refers to maintaining, using, and generating traces among software artefacts—i.e., triples comprised of a source artefact, a target artefact, and a trace link—to support software quality assurance. Due to the effort of manually discovering trace links and the variety of software artefacts, trace link discovery tools have been proposed. Among these is OntoTraceV2.0, an ontology-based automatic reasoning and Natural Language Processing (NLP) tool for trace link discovery. In this paper, we evaluate how OntoTraceV2.0 affects subjects’ efficiency, effectiveness, and satisfaction during trace link discovery. We conducted three quasi-experiments with 70 subjects in total. We asked subjects to discover trace links between a set of semi-structured software requirements in natural language—i.e., user stories—and a conceptual model—i.e., existence dependency graphs (EDGs)—with the support of OntoTraceV2.0 and without tool support. OntoTraceV2.0 increased subjects’ median precision compared to manual trace link discovery, with an average recall decrease of 7%. Despite this, OntoTraceV2.0 enabled subjects to discover trace links 1.41–2.55 times faster, indicating a significant increase in efficiency. Moreover, OntoTraceV2.0 positively affected subjects’ perceived usefulness, while ease of use and intention to use remain areas for improvement. Qualitative feedback highlighted the need for better guidance, automation, clearer benefits of long-term traceability, and improved user experience. To improve lower recall and enhance effectiveness while improving satisfaction, we propose an improved architecture for OntoTrace. We expect our experience to allow researchers and practitioners to devise new and better tools for trace link discovery automation.
