Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Regulatory Impact Analysis in Software Repositories: From Category-Driven Modeling to Evidence-Grounded Tool Support
Blekinge Institute of Technology, Faculty of Computing, Department of Software Engineering.ORCID iD: 0000-0003-4239-7838
fortiss GmbH.ORCID iD: 0000-0003-4142-6967
Technical University of Munich.
itestra GmbH.
Show others and affiliations
(English)Manuscript (preprint) (Other academic)
Abstract [en]

Context: Software systems operating in regulated domains must be adapted as legal and regulatory requirements evolve. While identifying regulatory changes is a first step, determining which parts of a software repository are affected remains a manual and knowledge-intensive activity.

Objectives: This paper investigates how regulatory changes can be systematically mapped to affected parts of software repositories, and how this analysis can be supported through an evidence-grounded tool.

Method: Following a Design Science Research approach, we conducted an empirical investigation in collaboration with an industrial partner. We devel-oped a category-driven impact model linking regulatory change categories topotentially affected repository artifact types and engineering consequences. The model was validated and refined with practitioners. We designed and implemented ReguLens, which was examined through a formative expert evaluation and refined. The version was evaluated using a controlled benchmark of 25 distinct historical cases from three regulation-dependent open-sourcerepositories.

Results: Practitioner validation revealed that the relevance and severity of software impacts depend on system context, limiting the use of regulatory change categories as deterministic predictors. The formative expert evaluation highlighted the need for expert interpretation when assessing potentially affected artifacts. In the controlled benchmark, ReguLens achieved 99.3%F1-score and 96.0% balanced accuracy for repository-level impact detection. Artifact localization was less effective: macro F1 reached 64.1% at file level and 60.6% at container level, but only 21.1% at symbol level.

Conclusion: Practitioners can benefit from ReguLens by using it to narrow the search space, identify plausible implementation entry points, and inspect repository evidence. However, incomplete localization shows that human review remains necessary to interpret the evidence and confirm the required software changes. The findings highlight both the potential and the limitations of AI-assisted support for regulatory impact analysis.

Keywords [en]
Requirements Engineering, Regulatory Change, Impact Analysis, Regulatory Compliance, Tool Evaluation, Large Language Models
National Category
Software Engineering
Research subject
Software Engineering
Identifiers
URN: urn:nbn:se:bth-30655DOI: 10.13140/RG.2.2.18114.95683OAI: oai:DiVA.org:bth-30655DiVA, id: diva2:2114050
Available from: 2026-10-29 Created: 2026-10-29 Last updated: 2026-10-29Bibliographically approved
In thesis
1. Automated Change Analysis in Regulatory Requirements Engineering
Open this publication in new window or tab >>Automated Change Analysis in Regulatory Requirements Engineering
2026 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

Background: Software-intensive systems in regulated domains must continuously adapt to evolving regulations. Regulatory changes may affect requirements, workflows, architecture, implementation, documentation, and testing. Analyzing these changes and their consequences is therefore an important part of maintaining regulatory compliance, yet it remains largely manual, knowledge-intensive, dependent on domain expertise, and prone to errors.

Problem: Existing research provides limited support for treating regulatory change analysis as an integrated software engineering activity, due to four main shortcomings:(1) limited empirical understanding of how regulatory requirements and changes are handled in industrial practice;(2) a lack of structured conceptualizations that distinguish different types of regulatory changes and support systematic reasoning about their engineering relevance;(3) limited tool support for detecting, classifying, and interpreting regulatory changes in realistic regulatory contexts; and(4) limited support for translating identified regulatory changes into impact assessments across software artifacts and engineering activities.Together, these shortcomings create a gap between identifying changes in regulatory documents and systematically determining their consequences for software systems. At the same time, recent advances in Generative AI, particularly Large Language Models (LLMs), provide new opportunities to bridge this gap through their ability to process and reason over large volumes of textual information. However, how effectively such capabilities can support regulatory change and impact analysis in software engineering remains insufficiently understood.

Goal: This cumulative thesis investigates the challenges software engineers face when analyzing evolving regulations and their impact on software systems, and the extent to which these challenges can be addressed through automated and, in particular, LLM-assisted approaches.

Method: The research follows an iterative Design Science Research approach. First, we synthesize existing research and investigate industrial practices to identify challenges in Regulatory Requirements Engineering. Second, we establish foundations for automation by deriving requirements for compliance automation systems and developing a taxonomy and conceptual model of regulatory changes. Third, we design and experimentally evaluate LLM-assisted methods for regulatory change analysis. Finally, we extend this support towards regulatory impact analysis by developing and evaluating a category-driven impact model and ReguLens, an evidence-grounded tool for translating detected regulatory changes into potentially affected software artifacts.

Results: Our research provides a consolidated understanding of Regulatory Requirements Engineering and empirical insights into industrial practices, challenges, and opportunities for automation. It establishes a structured set of engineering requirements for designing and evaluating human-centered compliance automation systems, alongside a taxonomy and conceptual model for systematically representing regulatory changes. The experimental evaluation shows that LLMs can support the detection and classification of recurring and structurally explicit regulatory changes, while performance decreases for narrative, weakly structured, or context-dependent changes, reinforcing the need for human involvement. The research further extends regulatory change analysis towards software impact assessment through a category-driven impact model and the ReguLens. The industrial evaluation provides first evidence that this approach can well support practitioners by providing structured starting points for implementation investigation, while also revealing limitations related to change granularity, the distinction between contextual conditions and implementation actions, and fine-grained localization of affected software artifacts.

Conclusion: The thesis establishes a systematic foundation for regulatory change and impact analysis and demonstrates how LLM-assisted and tool-supported approaches can augment human expertise within a structured engineering process. The findings position such approaches as decision support rather than autonomous compliance solutions, with practitioners retaining responsibility for interpreting regulatory changes, validating their impacts, and making implementation decisions.

Place, publisher, year, edition, pages
Karlskrona: Blekinge Tekniska Högskola, 2026. p. 200
Series
Blekinge Institute of Technology Doctoral Dissertation Series, ISSN 1653-2090 ; 2026:10
Keywords
Requirements Engineering, Regulatory Change Analysis, Regulatory Impact Analysis, Large Language Models, Design Science Research
National Category
Software Engineering
Identifiers
urn:nbn:se:bth-30665 (URN)978-91-7295-537-0 (ISBN)
Public defence
2026-12-10, J1630, Valhallavägen 1, Karlskrona, 13:00 (English)
Opponent
Supervisors
Available from: 2026-10-29 Created: 2026-10-29 Last updated: 2026-10-29Bibliographically approved

Open Access in DiVA

No full text in DiVA

Other links

Publisher's full texthttps://doi.org/10.13140/RG.2.2.18114.95683

Authority records

Elahidoost, ParisaFucci, DavideUnterkalmsteiner, MichaelMendez, Daniel

Search in DiVA

By author/editor
Elahidoost, ParisaVillamizar, HugoFucci, DavideUnterkalmsteiner, MichaelMendez, Daniel
By organisation
Department of Software Engineering
Software Engineering

Search outside of DiVA

GoogleGoogle Scholar

doi
urn-nbn

Altmetric score

doi
urn-nbn
Total: 18 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf