Automated Software Requirements Elicitation: A Systematic Mapping Study Review

Existing surveys on AI-based requirements elicitation are limited in scope: most focus on a single AI paradigm (e.g., machine learning or Natural Language Processing alone), omit textual-source taxonomies, lack cross-dimensional classification, and do not cover the recent surge in large language model (LLM) research. To address these gaps, this paper presents a systematic mapping study (SMS) of 76 peer-reviewed studies published between 2021 and 2025, identified following PRISMA 2020 guidelines from five major databases (IEEE Xplore, ACM, Scopus, SpringerLink, Web of Science). The selected studies are classified along four dimensions: AI techniques employed (machine learning, deep learning, transformer-based models, ontology-based reasoning, and LLMs), textual data sources (formal documentation, agile artifacts, conversational data, and hybrid corpora), research contribution types, and application areas within requirements engineering. The results reveal a decisive shift from traditional ML classifiers toward transformer-based and hybrid knowledge-enhanced architectures—including RAG-augmented pipelines, multi-agent LLM frameworks, and edge-cloud collaborative systems—with a growing emphasis on contextual understanding and partial automation of requirements generation. Requirements extraction and classification dominate current research (47.4%), while end-to-end automated generation remains limited to fewer than 25% of studies. Key open challenges include the absence of standardized benchmark datasets, limited model explainability, domain-specific overfitting, insufficient industrial validation, and neglect of non-functional requirements. Compared with prior surveys, this study offers a broader temporal scope, a multi-dimensional classification framework, and an integrated taxonomy of textual sources for AI-based elicitation that, to the best of our knowledge, has not previously been consolidated within a single systematic mapping study, providing actionable guidance for researchers and practitioners.Requirements elicitation is a foundational stage of the software engineering lifecycle in which stakeholder needs are identified and translated into system functionalities 1. Deficiencies at this early stage are a leading cause of project failures, cost overruns, and delayed deliveries 2. The process is inherently complex: communication barriers among diverse stakeholders, cognitive biases, human errors, and the challenge of capturing ambiguous or continuously evolving expectations all impede effective elicitation 3. Conventional techniques—structured interviews, facilitated workshops, and stakeholder surveys—remain predominantly manual, time-intensive, and heavily dependent on the expertise and interpretive judgment of requirements analysts 4. Consequently, these traditional approaches often yield inconsistent, incomplete, or ambiguous requirement specifications that propagate errors throughout the development lifecycle 5. Given these persistent limitations, the development of more efficient and accurate approaches to requirements elicitation has become a critical research priority for both the academic software engineering community and industrial practitioners 6. In recent years, advances in Artificial Intelligence (AI), Natural Language Processing (NLP), and Machine Learning (ML) have demonstrated considerable promise in addressing these fundamental limitations 7.Automated Software Requirements Elicitation (ASRE) leverages these converging capabilities to automatically identify, extract, analyze, and interpret requirements from diverse and heterogeneous data sources—including structured documentation, stakeholder communications, online user reviews, feedback repositories, and software issue tracking systems 8,9,10. By automating repetitive, labor-intensive analytical tasks, ASRE approaches reduce manual effort, minimize interpretation ambiguity, and enhance both the consistency and completeness of elicited requirements. Furthermore, automation supports continuous and iterative analysis, improving adaptability and responsiveness in agile and model-driven development environments 11, 12. Despite this demonstrated potential, automated requirements elicitation remains a rapidly evolving and fragmented research area characterized by diverse methodological approaches and inconsistent evaluation practices 13. Knowledge fragmentation persists, as existing studies predominantly address isolated objectives—such as requirement classification, prioritization, or validation—rather than comprehensive end-to-end elicitation solutions 5, 14 . Moreover, the absence of universally accepted evaluation metrics, standardized benchmark datasets, and reference frameworks significantly complicates the systematic assessment and comparison of automated elicitation techniques 15. This lack of methodological coherence creates substantial uncertainty for both researchers seeking to advance the field and practitioners attempting to adopt these technologies in industrial contexts 16. A comprehensive and systematic review is therefore urgently required to consolidate, synthesize, and critically analyze the current state of research. Several prior surveys have examined aspects of AI in requirements engineering; however, they differ in scope, classification depth, and temporal coverage. Table 1 systematically compares the present study with the most closely related reviews. Lim et al. 6 focused exclusively on data-driven elicitation techniques without distinguishing among AI paradigms. Sonbol et al. 17 concentrated on NLP-based text representations for RE tasks but did not cover deep learning or LLM-based approaches. Cai et al. 18 reviewed automatic elicitation from user-generated content but limited their scope to a single data-source category. Siddeshwar et al. 19 surveyed AI-enabled frameworks for elicitation but did not provide a textual-source taxonomy or cross-dimensional classification. Umar and Lano 20 offered a broad review of automated RE support but covered studies only up to 2023. In contrast, the present study spans 2021–2025, classifies 76 studies along four complementary dimensions (AI techniques, textual sources, contribution types, and application areas), and provides the first integrated taxonomy of textual inputs for AI-based elicitation. The present study conducts a Systematic Mapping Study (SMS) designed to comprehensively map and characterize the current research landscape in AI-based automated software requirements elicitation. The primary objectives are to identify prevailing research trends, systematically classify widely adopted methods and techniques, delineate key research domains, and expose persistent challenges and unresolved gaps. Through the systematic review of a broad corpus of 76 studies, this mapping study aims to uncover emerging thematic patterns and articulate actionable directions for future research. The findings contribute to a clearer understanding of the transformative role of automation within requirements engineering and provide valuable guidance for developing more effective, intelligent, and practically deployable elicitation tools. The motivation for this research stems from the increasing complexity and rapidly growing volume of unstructured textual data in modern software development projects, which renders manual extraction of requirements increasingly slow, inefficient, and error-prone. Although various AI techniques have been applied to automate this task, the existing research landscape lacks a unified and coherent perspective due to fragmentation across methods, datasets, and application domains.