PaperlayerView exact revision on arXiv

Source-linked AI summary

LLM-Assisted Workflow for Structural Difference Visualization in Evolving Software Requirements

Koi McFarland, Songhui Yue

arXiv:2609.28002v1cs.SEcs.HCcs.IR

TL;DR

Textual comparisons may not show how evolving requirements change semantically and structurally. The paper presents an LLM-assisted workflow for curating and comparing requirement graphs, with preliminary observations suggesting greater usefulness for complex changes than short, simple ones.

  • Problem

    Textual comparisons may expose surface updates without clearly showing structural changes in requirement concepts and relationships.

    Source: §I · p. 1

  • Method

    The workflow uses LLM-generated triples, human curation, saved graph snapshots, and side-by-side comparison with alignment, change encoding, and relation evidence inspection.

    Source: A. Workflow Overview · p. 1, B. Snapshot-based Comparison · p. 1

  • Results

    The preliminary assessment suggests the workflow is more useful for medium- and high-complexity changes, while direct textual comparison may suffice for short, simple changes.

    Source: §III · p. 2

  • Takeaways & Limitations

    The contribution is an integrated requirements-engineering workflow, rather than a new ontology-difference algorithm.

    Source: §IV · p. 2

    The prototype is better suited to selected short sections than multi-paragraph requirement documents; paragraph-by-paragraph comparison remains future work.

    Source: §IV · p. 2

Abstract

from arXiv · show

This paper presents an LLM-assisted workflow for visualizing structural differences in evolving software require- ments. Implemented in the OntologyWeb environment, the work- flow represents baseline and current requirements as triple-based semantic graphs and supports side-by-side comparison of curated graph snapshots. The comparison view aligns matched entities and uses visual encoding to highlight structural changes.

I. INTRODUCTION

Evolving requirements can change in meaning and relationships, not just wording, while textual comparison may obscure structural updates. The paper proposes an OntologyWeb workflow combining LLM-assisted structuring, human curation, and graph comparison for this use case.

I. INTRODUCTION

  • Textual comparisons may miss structural changes such as added responsibilities, removed dependencies, or changed relationships among requirement concepts.Understanding these changes relates to requirements traceability and change impact analysis.Source: §I · p. 1
  • LLM-generated triples and ontology structures require human examination and validation, preservation as comparable versions, and visualization for structural change analysis.Triple-based representations let analysts inspect concepts and relations more explicitly than plain text.Source: §I · p. 1
  • OntologyWeb targets requirement-version comparison by combining LLM-assisted generation, human curation, snapshot selection, evidence inspection, and synchronized side-by-side graph views.Matched entities are placed in similar visual positions across baseline and current snapshots.Source: §I · p. 1

A. Workflow Overview

The OntologyWeb workflow turns two requirement versions into curated, comparable graph snapshots through a four-step process. It supports inspection of common, added, removed, and modified structures.

A. Workflow Overview

  • The workflow processes baseline and current requirements with the same LLM prompt and model settings to generate candidate triples, which users can edit, add to, or remove.Users save the curated graph versions as snapshots, then select two snapshots for side-by-side comparison.Source: A. Workflow Overview · p. 1
  • The workflow connects candidate-triple generation, human curation, snapshot saving, and comparison of selected snapshots.The workflow is implemented in OntologyWeb, a web-based visualization-oriented tool supporting human-in-the-loop curation of LLM-generated triples and ontology fragments.Source: A. Workflow Overview · p. 1
  • Comparing snapshots supports structural-difference analysis of common, added, removed, and modified structures across requirement versions.Source: A. Workflow Overview · p. 1

B. Snapshot-based Comparison

The comparison view presents two curated graph snapshots from different versions of the same requirement. Its alignment, change encoding, and relation evidence support inspection of structural differences.

B. Snapshot-based Comparison

  • The view compares two curated graph snapshots from different versions of the same requirement, aligning shared concepts and highlighting structural differences.Figure 1(b) illustrates an example comparison view.Source: B. Snapshot-based Comparison · p. 1
  • Matched nodes appear in approximately similar positions across baseline and current graphs, while colored node or edge borders indicate additions, removals, and modifications.Source: B. Snapshot-based Comparison · p. 1
  • Hovering over relations lets users inspect evidence text and trace each generated relation to its original requirement text.Source: B. Snapshot-based Comparison · p. 1

C. Implementation and Demonstration

The implementation uses a centralized JSON representation for graph processing and rendering. Figure 1 outlines the workflow and comparison example, and a short video demonstrates the tool.

C. Implementation and Demonstration

III. PRELIMINARY OBSERVATIONS

A preliminary qualitative assessment used real-world requirement documents with multiple versions to examine structural-difference inspection. Observations suggest the workflow suits medium- and high-complexity changes better than short, simple ones.

III. PRELIMINARY OBSERVATIONS

  • The exploratory assessment compared baseline and current versions of real-world requirement documents, focusing on inspection of added, removed, and modified relations.Each document had at least two versions; the assessment processed version pairs through the workflow and compared their resulting graph structures.Source: §III · p. 2
  • The workflow appears more useful for medium- and high-complexity changes, while direct textual comparison may suffice for short, simple changes.For complex requirements, the side-by-side view appeared to make relations, dependencies, and changed structures more explicit, though it may require additional inspection effort for short paragraphs.Source: §III · p. 2
  • Future evaluation will measure task completion time, structural-difference identification accuracy, user confidence, and subjective usability.Source: §III · p. 2

IV. DISCUSSION AND CONCLUSION

The paper contributes an integrated workflow for requirement-version comparison, rather than a new ontology-difference algorithm. The prototype is currently better suited to selected short sections than to multi-paragraph documents.

IV. DISCUSSION AND CONCLUSION

  • The prototype is more suitable for comparing selected short sections than multi-paragraph requirements; paragraph-by-paragraph comparison remains a future direction.Source: §IV · p. 2
  • The primary contribution is an integrated requirement-engineering workflow connecting LLM-assisted requirement structuring with interactive graph comparison, not a new ontology-difference algorithm.Source: §IV · p. 2
  • The workflow combines human curation of requirement triples, snapshot-based graph comparison, and synchronized side-by-side visualization with color encoding, evidence inspection, and layout alignment.Source: §IV · p. 2
Loading 2609.28002v1…