Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

LLMREI: Automating Requirements Elicitation Interviews with LLMs

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read LLMREI chatbot runs requirements interviews with near-human error rates.

desk verdict A reproducible feasibility study of an LLM interview bot, but the headline claim of parity with human interviewers rests on a confounded comparison. read the letter →

arxiv 2507.02564 v1 pith:VKOYXGTR submitted 2025-07-03 cs.SE

Alexander Korn , Samuel Gorsch , Andreas Vogelsang This is my paper · ORCID
classification cs.SE
keywords requirementselicitationinterviewautomationlargelanguagemodelschatbotpromptengineeringzero-shotpromptingleast-to-most
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Requirements elicitation interviews depend on skilled analysts and are hard to scale, so the paper asks whether a large language model can run those interviews itself. It introduces LLMREI, a chat bot built on GPT-4o with two prompt designs: a short zero-shot instruction and a long least-to-most style prompt that encodes published interview guidelines. Across 33 simulated stakeholder interviews in two scenarios, the bot kept roughly the same rate of common interviewer mistakes as trained human interviewers from a prior study, and it fully or partially elicited up to 73.7% of the requirements in the scenario descriptions. The authors also show that about half of the bot's questions were context-dependent, meaning it followed up on the stakeholder's own answers rather than reading from a fixed list. If this holds, early-phase interviews with many stakeholders could be largely delegated to a chatbot, with human analysts focusing on sensitive or complex follow-ups.

What carries the argument

The load-bearing object is the system prompt. LLMREI runs on GPT-4o with one of two prompts. The short prompt is a three-sentence zero-shot instruction: act as an interviewer called LLMREI, bombard the stakeholder with business and project questions, and ask one question at a time, or two if closely related. The long prompt is a least-to-most style prompt that adds a role description, a five-step interview cookbook, question pathways, error-handling rules, and examples, refined iteratively with mistake categories and interview guidelines from the literature. The evaluation machinery consists of a Likert questionnaire mapped one-to-one to common interview mistakes, manual coding of every bot question into four context-dependence categories, and recall computed against a manually built ground-truth requirement list per scenario.

What would settle it

Have the same panel of raters score chatbot-conducted and human-conducted interviews of the same scenario with the same questionnaire; if humans score significantly better on the common-mistake items, the similar-error claim fails. A second check: compute requirement recall against a ground-truth list created before the interviews and kept blind to the bot's prompts; if recall drops well below 73.7%, the effectiveness claim is not reproducible.

Watch

Extended reading notes

Core claim

The central claim is that LLMREI demonstrates a prompt-engineering path to automating requirements elicitation interviews. The short prompt alone was enough to make GPT-4o behave like an interviewer, asking one or two related questions, probing, and switching topics smoothly. The long prompt, refined through iterative testing against published mistake categories, reduced errors further in question formulation, ordering, and closing summaries. On effectiveness, the short prompt fully elicited 60.94% of the scenario requirements and, counting partially elicited items, reached 73.7% of the ground-truth set; the long prompt covered a slightly smaller share. Both versions produced a comparable share of context-deepening and context-enhancing questions, and the bot handled non-native speakers and a mid-interview switch to German without derailment. The paper concludes that LLMs can match trained human interviewers on common mistakes while still missing a meaningful slice of requirements, and that full automation is not yet achieved.

Load-bearing premise

The central comparison assumes that the questionnaire ratings from a separate earlier human-interviewer study can be directly compared with the chatbot ratings, even though the raters, interview contexts, and assessment procedures differed and no statistical adjustment was made.

Editorial extensions

If this is right

  • Requirements teams could run the bot as a first-pass interviewer across many stakeholders and expect over 70% of the scenario requirements to surface before a human analyst follows up.
  • The long prompt's explicit instructions materially change behavior: reviewers rated it better on final summaries and question length, so prompt complexity is a usable control knob.
  • A single GPT-4o conversation can adapt to language switches, sparse answers, and users who want to end early, without custom integration.
  • Fine-tuning on real interview transcripts is not a quick win here: all fine-tuned variants performed worse than the two prompted versions and were dropped from the evaluation.
  • Because the bot follows prompt instructions reliably, the remaining gap in requirement coverage is plausibly addressable by further prompt refinement rather than by changing the model.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The similar-errors comparison rests on ratings from a separate earlier study with different raters and settings; a same-rater head-to-head trial could shift the error gap in either direction.
  • The scenario ground truth is known to the evaluators, so the 73.7% recall may upper-bound real-world performance, where the full set of requirements is not enumerable in advance.
  • The bot's context-enhancing questions occasionally steer the conversation toward features the stakeholder did not mention, so teams wanting neutral elicitation might want a control on how often the bot proposes ideas.
  • Hallucinated figures and a request for a personal email address appeared in individual interviews, so production use would likely require a guardrail layer that blocks out-of-scope claims and personal-data requests.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper introduces LLMREI, a GPT-4o-based chatbot designed to conduct requirements elicitation interviews with minimal human intervention. It compares two prompting strategies—a concise zero-shot prompt (LLMREI-short) and a longer, iteratively refined prompt (LLMREI-long)—across 33 simulated stakeholder interviews in two scenarios. The evaluation addresses three research questions: the frequency of common interviewer mistakes (RQ1, assessed with a questionnaire adapted from Bano et al. and Ferrari et al. and compared to a human-interviewer baseline from Ferrari et al.), the proportion of scenario requirements elicited (RQ2, against author-created ground-truth lists), and the extent to which the bot's questions adapt to conversational context (RQ3, via manual question categorization). The authors report that LLMREI makes a similar number of common mistakes as human interviewers, elicits up to 73.7% of requirements, and generates many context-dependent questions. A fine-tuning approach using GPT-3.5 was attempted and abandoned.

Significance. If the central claims hold, the paper would provide a valuable empirical data point on whether LLM-based chatbots can conduct useful requirements interviews with limited human involvement. The availability of a replication package containing raw interview data, ground-truth labels, and analysis code is a clear strength. However, the headline human-equivalence claim for RQ1 is currently not established: it rests on an uncontrolled comparison with a separate study, with different raters, different procedures, and no statistical tests. RQ2 and RQ3 offer useful descriptive evidence but would benefit from stronger reliability reporting. The paper is within the scope of the venue and the underlying idea is promising, but the load-bearing comparison needs substantial rework.

major comments (4)
  1. [Section IV.C and Figures 2-3] The central claim that LLMREI "makes a similar number of common mistakes" compared to human interviewers is supported only by a descriptive comparison between two different studies. The LLMREI ratings come from questionnaires completed by the 33 participants in the new experiment, while the human baseline comes from 18 student-led interviews selected from Ferrari et al.'s study, rated by different people under different conditions. No statistical test (e.g., per-item Mann-Whitney or a mixed-effects model), rater calibration, or adjustment for rater leniency is reported. The figures show percentage distributions of Likert responses, not error rates with uncertainty, so the visible differences in items such as "No final summary" or "Not asking about feature prioritization" cannot be attributed to the interviewer rather than the raters. The paper itself acknowledges in Section IV.D that the sample size "limits statistical significance," yet the abstract and Section I state the similarity as a finding. To support RQ1, the authors should either collect a human-interviewer baseline under the same protocol or substantially temper the claim and provide an appropriate statistical analysis of the available data.
  2. [Section III.B and Section IV.C] The long prompt was iteratively refined by categorizing transcripts with Bano et al.'s mistake framework and by using Ferrari et al.'s interview guidelines, and the RQ1 questionnaire was then derived from the same Bano/Ferrari mistake taxonomy. This creates a partial circularity: the long prompt was tuned to avoid the very mistake categories that later serve as the evaluation items. The paper does not report which mistake categories were used during prompt refinement and which were used in the questionnaire, so the degree of overlap cannot be assessed. At minimum, the manuscript should disclose this overlap and discuss its implications; ideally, the prompt should be evaluated on held-out mistake categories or with independent raters who are blind to the prompt-development process. If the long prompt's perceived superiority is partly an artifact of prompt tuning on the evaluation taxonomy, the general claim about reducing interview errors is weakened.
  3. [Section IV.B, Section IV.C, and Section V (RQ2)] The recall measure for RQ2 is based on author-created ground-truth lists of only 8 and 12 requirements per scenario, and the two reviewers who manually labelled elicited requirements are discussed only as reaching consensus; no inter-rater reliability statistic (e.g., Cohen's kappa) is reported. In addition, the Discussion states that "There was no significant difference in the number of elicited requirements between the long and short prompts," but no significance test is reported anywhere in the evaluation. The paper should report the labelling agreement, provide confidence intervals or a proper test for the long-vs-short recall comparison, and avoid unsupported claims of non-significance.
  4. [Section IV.C and Section IV.E (RQ3)] The four-category question classification (context-independent, parameterized, context-deepening, context-enhancing) is performed manually, yet no inter-rater reliability is reported. Given the centrality of the "adaptability" claim and the modest differences in Table II, a second rater and an agreement measure (e.g., Cohen's kappa) are needed to establish that the categorization is reliable. Without this, the RQ3 conclusion that roughly half of the questions are context-dependent or context-enhancing remains a single-rater impression rather than a measured result.
minor comments (5)
  1. [Section III.B and Contributions] The paper explicitly states that the long-prompt approach is a "session-level refinement process" rather than least-to-most prompting in the strict sense, yet the abstract and the contribution bullet list "least-to-most prompting" as one of the two tested approaches. Rename this approach to "iterative prompt refinement" or provide a clear justification for the least-to-most label.
  2. [Section IV.C] There is a typo, "elcitation," in the paragraph introducing Ferrari et al.'s study. Also, in the RQ3 definition, "the interviewer's responses" should be "the interviewee's responses."
  3. [Section IV.B] The criteria for what counts as a requirement in the ground truth are not stated. The authors should specify how the 12 (Ski) and 8 (Salon) requirements were derived and what granularity they reflect, since the recall percentages depend entirely on this list.
  4. [Section IV.C] The questionnaire adaptation omitted questions "not applicable to an interview bot environment," but the manuscript does not list which questions were omitted. Listing them would help readers assess how the comparison with the human baseline is affected.
  5. [Figure 5] The percentage values inside the cells are difficult to read, and the figure does not convey uncertainty. A table with 95% confidence intervals for each requirement would be more informative.

Circularity Check

1 steps flagged · score 4.0 of 10

RQ1's 'similar number of errors' is partially self-fulfilling: the LLMREI-long prompt was iteratively tuned on Bano et al.'s mistake taxonomy, then scored with a questionnaire derived from that same taxonomy.

  1. fitted input called prediction [Section III.B (least-to-most prompting development) and Section IV.C (RQ1 questionnaire construction)]
    "We followed an iterative development cycle: after testing each version in stakeholder interviews, we analyzed transcripts and categorized mistakes using Bano et al.'s framework [25]. Revisions were then informed by Ferrari et al.'s requirements interview guidelines [26] ... To address RQ1, we focused on common mistake types identified in requirements interviews to create our questionnaire. These mistake types are derived from the study of Bano et al. [25]."

    The LLMREI-long prompt was explicitly refined by categorizing its own transcripts with Bano et al.'s mistake framework and by applying Ferrari et al.'s guidelines, which operationalize the same mistake categories. The RQ1 evaluation then scores the bot with a questionnaire 'derived from the study of Bano et al.,' with one question per mistake type. Thus the long prompt's low error scores on RQ1 largely measure how well the prompt was optimized against the very taxonomy used as the test, rather than providing an independent estimate of interview quality. The paper's headline claim that LLMREI 'makes a similar number of common mistakes' as human interviewers is therefore partly self-fulfilling for LLMREI-long.

full rationale

This paper is an empirical evaluation, not a formal derivation, so no equation reduces to its inputs. The main circularity concern is the coupling between prompt development and evaluation: LLMREI-long was iteratively improved by categorizing mistakes with Bano et al.'s framework, and the RQ1 questionnaire was built from that same framework. This makes the long prompt's error scores partly a reflection of fitting to the evaluation instrument. The comparison to human interviewers from Ferrari et al. also involves different raters and contexts with no calibration, but that is a validity threat rather than circularity. RQ2, using an author-created ground truth, and RQ3, based on manual question categorization, are independent of the Bano taxonomy and provide separate evidence for the paper's other claims. Given that the central error-equivalence claim retains some independent support, especially from the untuned short prompt, a moderate score of 4 is appropriate.

Assumptions & free parameters 2 free parameters · 4 assumptions · 1 invented entities

The central claim rests on several domain assumptions about the validity of simulated stakeholder interviews and the transferability of a human-oriented mistake framework to LLMs. The prompt for LLMREI-long was iteratively refined using the same mistake framework later used to evaluate it, which makes the prompt a fitted artifact rather than a fixed intervention.

free parameters (2)
  • LLMREI-long prompt content = Iteratively refined; full text in replication package
    Each iteration adjusted the prompt based on mistake categories from Bano et al.'s framework, and the final version was then evaluated with a questionnaire derived from the same framework.
  • Ground truth requirement lists = 12 requirements (ski resort), 8 (hair salon)
    The authors derived these lists from the scenarios by consensus; there is no external validation that they are complete or unambiguous.
assumptions (4)
  • domain assumption Bano et al.'s mistake framework is applicable to an LLM interviewer
    The evaluation questionnaire is built from that framework, but the framework was designed for human interviewers; the authors themselves note this in Section IV.D.
  • domain assumption Student participants acting as stakeholders provide a valid proxy for real stakeholders
    All participants are computer science students, not actual stakeholders; the authors flag this in Section IV.D.
  • domain assumption Interviewee self-ratings are comparable to assessments of human interviewers made in a different study
    The human baseline data come from Ferrari et al., with different raters and procedures; the paper applies no correction for this difference (Section IV.C).
  • ad hoc to paper The iterative session-level prompt refinement is an instance of least-to-most prompting
    The paper admits it is not least-to-most in the strict sense of Zhou et al., but a session-level refinement process (Section III.B).
invented entities (1)
  • LLMREI chatbot independent evidence
    purpose: Conducts automated requirements elicitation interviews with stakeholders.
    The behavior is specified by the shipped prompts and can be reproduced with GPT-4o; the replication package contains the prompt texts and interview transcripts.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LLMREI: Automating Requirements Elicitation Interviews with LLMs." pith.science (2026). https://pith.science/paper/VKOYXGTR

@misc{pith2026250702564,
  author       = {Pith},
  title        = {Pith review of: LLMREI: Automating Requirements Elicitation Interviews with LLMs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VKOYXGTR}},
  note         = {Machine review of arXiv:2507.02564}
}
read the original abstract

Requirements elicitation interviews are crucial for gathering system requirements but heavily depend on skilled analysts, making them resource-intensive, susceptible to human biases, and prone to miscommunication. Recent advancements in Large Language Models present new opportunities for automating parts of this process. This study introduces LLMREI, a chat bot designed to conduct requirements elicitation interviews with minimal human intervention, aiming to reduce common interviewer errors and improve the scalability of requirements elicitation. We explored two main approaches, zero-shot prompting and least-to-most prompting, to optimize LLMREI for requirements elicitation and evaluated its performance in 33 simulated stakeholder interviews. A third approach, fine-tuning, was initially considered but abandoned due to poor performance in preliminary trials. Our study assesses the chat bot's effectiveness in three key areas: minimizing common interview errors, extracting relevant requirements, and adapting its questioning based on interview context and user responses. Our findings indicate that LLMREI makes a similar number of errors compared to human interviewers, is capable of extracting a large portion of requirements, and demonstrates a notable ability to generate highly context-dependent questions. We envision the greatest benefit of LLMREI in automating interviews with a large number of stakeholders.

Figures

Figures reproduced from arXiv: 2507.02564 by the authors.

Figure 1
Figure 1. Conversation excerpt from an interview conducted by the LLMREI [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Evaluation of common mistakes made by the [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 4
Figure 4. Ratio of fully and partially elicited requirements (i.e., [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figures from the paper (1 more)
Figure 6
Figure 6. Figure 6: Conversation excerpt from an interview conducted by the LLMREI [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Automated Alignment between Elicitation Interviews and Requirements

    cs.CL 2025-10 conditional novelty 5.0 of 10

    TEXT2STORIES formalizes chunk–story alignment with correctness and completeness metrics; an LLM judge reaches 0.86 macro-F1 on manually labeled pairs.

Reference graph

Works this paper leans on

33 extracted references · 28 canonical work pages · cited by 1 Pith paper

  1. [1]

    The role of requirement engineering in software development life cycle,

    A. Chakraborty, M. K. Baowaly, A. Arefin, and A. N. Bahar, “The role of requirement engineering in software development life cycle,” Journal of emerging trends in computing and information sciences , vol. 3, no. 5, 2012

    work page 2012

  2. [2]

    Requirements elicitation: A survey of tech- niques, approaches, and tools,

    D. Zowghi and C. Coulin, “Requirements elicitation: A survey of tech- niques, approaches, and tools,” in Engineering and Managing Software Requirements. Springer-Verlag, 2005, pp. 19–46

    work page 2005

  3. [3]

    Requirements elicitation tech- niques: a systematic literature review based on the maturity of the techniques,

    C. Pacheco, I. García, and M. Reyes, “Requirements elicitation tech- niques: a systematic literature review based on the maturity of the techniques,” IET Software, vol. 12, no. 4, pp. 365–378, 2018

    work page 2018

  4. [4]

    Ambiguity and tacit knowledge in requirements elicitation interviews,

    A. Ferrari, P. Spoletini, and S. Gnesi, “Ambiguity and tacit knowledge in requirements elicitation interviews,” Requirements Engineering, vol. 21, no. 3, pp. 333–355, 2016

    work page 2016

  5. [5]

    On cognitive biases in re- quirements elicitation,

    A. Zalewski, K. Borowa, and D. Kowalski, “On cognitive biases in re- quirements elicitation,” in Integrating Research and Practice in Software Engineering, S. Jarzabek, A. Poniszewska-Mara ´nda, and L. Madeyski, Eds. Cham: Springer International Publishing, 2020, pp. 111–123

    work page 2020

  6. [6]

    Automatic requirements en- gineering model using goal-oriented modelling with text pre-processing technique,

    R. Delima, R. Wardoyo, and K. Mustofa, “Automatic requirements en- gineering model using goal-oriented modelling with text pre-processing technique,” in 6th International Conference on Informatics and Com- puting (ICIC), 2021, pp. 1–8

    work page 2021

  7. [7]

    Artificial intelligence in software requirements engineering: State-of-the-art,

    K. Liu, S. Reddivari, and K. Reddivari, “Artificial intelligence in software requirements engineering: State-of-the-art,” in 2022 IEEE 23rd International Conference on Information Reuse and Integration for Data Science (IRI), 2022, pp. 106–111

    work page 2022

  8. [8]

    A survey of large language models,

    W. X. Zhao, K. Zhou, J. Li, T. Tang, X. Wang, Y . Hou, Y . Min, B. Zhang, J. Zhang, Z. Dong, Y . Du, C. Yang, Y . Chen, Z. Chen, J. Jiang, R. Ren, Y . Li, X. Tang, Z. Liu, P. Liu, J.-Y . Nie, and J.-R. Wen, “A survey of large language models,” 2024. [Online]. Available: https://arxiv.org/abs/2303.18223

    arXiv 2024

Show all 33 references
  1. [9]

    Using large language models for natural language processing tasks in requirements engineering: A systematic guideline,

    A. V ogelsang and J. Fischbach, “Using large language models for natural language processing tasks in requirements engineering: A systematic guideline,” in Handbook of Natural Language Processing for Require- ments Engineering, A. Ferrari and G. Deshpande, Eds. Cham: Springer ...

    2025

  2. [10]

    Advancing requirements engineering through generative AI: Assessing the role of LLMs,

    C. Arora, J. Grundy, and M. Abdelrazek, “Advancing requirements engineering through generative AI: Assessing the role of LLMs,” in Generative AI for Effective Software Development . Springer Nature Switzerland, 2024, pp. 129–148

    2024

  3. [11]

    Use of expert system in requirements engineering process: A systematic literature review,

    B. Haq, M. Nadeem, I. Ali, K. Ali, M. Raza, and M. U. Rehmanr, “Use of expert system in requirements engineering process: A systematic literature review,” in UK/ China Emerging Technologies (UCET) , 2019, pp. 1–5

    2019

  4. [12]

    Domain specialization as the key to make large language models disruptive: A comprehensive survey,

    C. Ling, X. Zhao, J. Lu, C. Deng, C. Zheng, J. Wang, T. Chowdhury, Y . Li, H. Cui, X. Zhang, T. Zhao, A. Panalkar, D. Mehta, S. Pasquali, W. Cheng, H. Wang, Y . Liu, Z. Chen, H. Chen, C. White, Q. Gu, J. Pei, C. Yang, and L. Zhao, “Domain specialization as the key to make larg...

  5. [13]

    Issues in requirements elicitation,

    M. Christel and K. C. Kang, “Issues in requirements elicitation,” 1992

    1992

  6. [14]

    Requirements elicitation: Issues and chal- lenges,

    S. Sharma and S. K. Pandey, “Requirements elicitation: Issues and chal- lenges,” in 2014 International Conference on Computing for Sustainable Global Development (INDIACom) , 2014, pp. 151–155

    2014

  7. [15]

    The role of domain knowledge in requirements elicitation via interviews: an exploratory study,

    I. Hadar, P. Soffer, and K. Kenzi, “The role of domain knowledge in requirements elicitation via interviews: an exploratory study,” Require- ments Engineering, vol. 19, no. 2, pp. 143–159, 2014

    2014

  8. [16]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. u. Kaiser, and I. Polosukhin, “Attention is all you need,” in Advances in Neural Information Processing Systems , I. Guyon, U. V . Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garn...

    2017

  9. [17]

    Large language models for software engi- neering: A systematic literature review,

    X. Hou, Y . Zhao, Y . Liu, Z. Yang, K. Wang, L. Li, X. Luo, D. Lo, J. Grundy, and H. Wang, “Large language models for software engi- neering: A systematic literature review,” ACM Transactions on Software Engineering and Methodology , vol. 33, no. 8, pp. 1–79, 2024

    2024

  10. [18]

    Large language models for software engineering: Survey and open problems,

    A. Fan, B. Gokkaya, M. Harman, M. Lyubarskiy, S. Sengupta, S. Yoo, and J. M. Zhang, “Large language models for software engineering: Survey and open problems,” in IEEE/ACM International Conference on Software Engineering: Future of Software Engineering (ICSE-FoSE). IEEE, 2023,...

    2023

  11. [19]

    A transformer-based approach for abstractive summarization of requirements from obligations in software engineering contracts,

    C. Jain, P. R. Anish, A. Singh, and S. Ghaisas, “A transformer-based approach for abstractive summarization of requirements from obligations in software engineering contracts,” in IEEE 31st International Require- ments Engineering Conference (RE) . IEEE, 2023, pp. 169–179

    2023

  12. [20]

    Prompts matter: Insights and strategies for prompt engineering in automated soft- ware traceability,

    A. D. Rodriguez, K. R. Dearstyne, and J. Cleland-Huang, “Prompts matter: Insights and strategies for prompt engineering in automated soft- ware traceability,” in 31st IEEE International Requirements Engineering Conference Workshops (REW), K. Schneider, F. Dalpiaz, and J. Horko...

    2023

  13. [21]

    On the impact of requirements smells in prompts: The case of automated traceability,

    A. V ogelsang, A. Korn, G. Broccia, A. Ferrari, J. Fischbach, and C. Arora, “On the impact of requirements smells in prompts: The case of automated traceability,” in IEEE/ACM International Conference on Software Engineering (ICSE) , 2025

    2025

  14. [22]

    Zadenoori, L

    M. Zadenoori, L. Zhao, W. Alhoshan, and A. Ferrari, Automatic Prompt Engineering: the Case of Requirements Classification . Springer Inter- national Publishing, 2025

    2025

  15. [23]

    Model generation with llms: From requirements to UML sequence diagrams,

    A. Ferrari, S. Abualhaija, and C. Arora, “Model generation with llms: From requirements to UML sequence diagrams,” in IEEE 32nd Interna- tional Requirements Engineering Conference Workshops (REW), vol. 24. IEEE, 2024, p. 291–300

    2024

  16. [24]

    Least-to- most prompting enables complex reasoning in large language models,

    D. Zhou, N. Schärli, L. Hou, J. Wei, N. Scales, X. Wang, D. Schuurmans, C. Cui, O. Bousquet, Q. Le, and E. Chi, “Least-to- most prompting enables complex reasoning in large language models,”

  17. [25]

    Teaching requirements elicitation interviews: an empirical study of learning from mistakes,

    M. Bano, D. Zowghi, A. Ferrari, P. Spoletini, and B. Donati, “Teaching requirements elicitation interviews: an empirical study of learning from mistakes,” Requirements Engineering, vol. 24, no. 3, pp. 259–289, 2019

    2019

  18. [26]

    SaPeer and Revers- eSaPeer: teaching requirements elicitation interviews with role-playing and role reversal,

    A. Ferrari, P. Spoletini, M. Bano, and D. Zowghi, “SaPeer and Revers- eSaPeer: teaching requirements elicitation interviews with role-playing and role reversal,” Requirements Engineering, vol. 25, no. 4, pp. 417– 438, 2020

    2020

  19. [27]

    Learning requirements elicitation interviews with role-playing, self-assessment and peer-review,

    ——, “Learning requirements elicitation interviews with role-playing, self-assessment and peer-review,” in IEEE 27th International Require- ments Engineering Conference (RE) , 2019, pp. 28–39

    2019

  20. [28]

    Whisperx: Time-accurate speech transcription of long-form audio,

    M. Bain, J. Huh, T. Han, and A. Zisserman, “Whisperx: Time-accurate speech transcription of long-form audio,” in INTERSPEECH 2023 . ISCA, 2023

    2023

  21. [29]

    Robust speech recognition via large-scale weak super- vision,

    A. Radford, J. W. Kim, T. Xu, G. Brockman, C. Mcleavey, and I. Sutskever, “Robust speech recognition via large-scale weak super- vision,” in 40th International Conference on Machine Learning , ser. Proceedings of Machine Learning Research, A. Krause, E. Brunskill, K. Cho, B. E...

    2023

  22. [30]

    Preparing your dataset

    OpenAI, “Preparing your dataset.” [Online]. Available: https://platform. openai.com/docs/guides/fine-tuning/preparing-your-dataset

  23. [31]

    Intelligent chatbot for requirements elicitation and classification,

    C. S. Rajender Kumar Surana, Shriya, D. B. Gupta, and S. P. Shankar, “Intelligent chatbot for requirements elicitation and classification,” in 4th International Conference on Recent Trends on Electronics, Information, Communication & Technology (RTEICT) , 2019, pp. 866–870

    2019

  24. [2023]

    Available: https://arxiv.org/abs/2205.10625

    [Online]. Available: https://arxiv.org/abs/2205.10625

    arXiv

  25. [2024]

    Available: https://arxiv.org/abs/2305.18703

    [Online]. Available: https://arxiv.org/abs/2305.18703

    arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.