Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Designing NLP-based solutions for requirements variability management: experiences from a design science study at Visma

Parisa Elahidoost, Michael Unterkalmsteiner, Davide Fucci, Peter Liljenberg, Jannik Fischbach

TL;DR

The paper tackles variability management in contract-derived requirements by applying a design-science approach in an industry–academic collaboration with Visma. Through five iterative NLP-driven cycles—ranging from heatmap-based feasibility to topic modeling and sentence-level embeddings—the study demonstrates how progressively concrete artifacts and prototypes can enhance traceability of clause configurations. Key findings show that full automation is unattainable in regulatory texts, but partial automation combined with expert input improves efficiency and decision support. The work offers practical lessons for design science in industry settings and suggests a generalizable approach to variability and traceability in contract-based requirements across domains.

Abstract

Context and motivation: In this industry-academia collaborative project, a team of researchers, supported by a software architect, business analyst, and test engineer explored the challenges of requirement variability in a large business software development company. Question/problem: Following the design science paradigm, we studied the problem of requirements analysis and tracing in the context of contractual documents, with a specific focus on managing requirements variability. This paper reports on the lessons learned from that experience, highlighting the strategies and insights gained in the realm of requirements variability management. Principal ideas/results: This experience report outlines the insights gained from applying design science in requirements engineering research in industry. We show and evaluate various strategies to tackle the issue of requirement variability. Contribution: We report on the iterations and how the solution development evolved in parallel with problem understanding. From this process, we derive five key lessons learned to highlight the effectiveness of design science in exploring solutions for requirement variability in contract-based environments.

Designing NLP-based solutions for requirements variability management: experiences from a design science study at Visma

TL;DR

The paper tackles variability management in contract-derived requirements by applying a design-science approach in an industry–academic collaboration with Visma. Through five iterative NLP-driven cycles—ranging from heatmap-based feasibility to topic modeling and sentence-level embeddings—the study demonstrates how progressively concrete artifacts and prototypes can enhance traceability of clause configurations. Key findings show that full automation is unattainable in regulatory texts, but partial automation combined with expert input improves efficiency and decision support. The work offers practical lessons for design science in industry settings and suggests a generalizable approach to variability and traceability in contract-based requirements across domains.

Abstract

Context and motivation: In this industry-academia collaborative project, a team of researchers, supported by a software architect, business analyst, and test engineer explored the challenges of requirement variability in a large business software development company. Question/problem: Following the design science paradigm, we studied the problem of requirements analysis and tracing in the context of contractual documents, with a specific focus on managing requirements variability. This paper reports on the lessons learned from that experience, highlighting the strategies and insights gained in the realm of requirements variability management. Principal ideas/results: This experience report outlines the insights gained from applying design science in requirements engineering research in industry. We show and evaluate various strategies to tackle the issue of requirement variability. Contribution: We report on the iterations and how the solution development evolved in parallel with problem understanding. From this process, we derive five key lessons learned to highlight the effectiveness of design science in exploring solutions for requirement variability in contract-based environments.
Paper Structure (10 sections, 4 figures, 1 table)

This paper contains 10 sections, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Research Methodology
  4. Design Iterations and Results
  5. Iteration One
  6. Iteration Two
  7. Iteration Three
  8. Iteration Four
  9. Iteration Five
  10. Conclusion

Figures (4)

  • Figure 1: Visualization of term matching frequencies in feature configurations (rows) and pages in a CLA (columns). Ground truth marked with a "T".
  • Figure 2: Graphical depiction of the proposed method - Iterations three, four and five
  • Figure 3: User task flow diagram for CLA analysis
  • Figure 4: Snippet of the result