AI IN SOFTWARE ENGINEERING
UCGen: Leveraging LLMs for Automated Generation of Use Case Textual Descriptions from Requirements Specification
UCGen leverages LLMs to automate the generation and validation of high-quality use case textual descriptions from natural language requirements, addressing challenges of manual generation like labor-intensiveness and error-proneness. It uses a human-in-the-loop (HITL) approach with tailored prompting strategies (zero-shot, few-shot, chain-of-thought, rule-guided chaining) for goal identification, actor extraction, scenario generation, and verification. Evaluation on ten Public Requirements (PURE) datasets shows UCGen produces more complete, correct, and less redundant use cases than human-derived ones, with HITL further enhancing quality. This domain-agnostic approach improves efficiency and quality in requirements engineering.
Executive Impact & Key Metrics
Dive into the measurable improvements and performance benchmarks achieved by leveraging advanced AI in requirements engineering.
0
Improvement in Correctness (Post-HITL)
0
Redundancy Increase (Post-HITL due to added scenarios)
0
Initial Correctness (Pre-HITL)
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Introduction
Proposed Approach
Evaluation and Analysis
Discussions
Conclusions
UCGen Pipeline Overview
Problem Specification (NL Text)
→
LLM (Generation)
→
Generated Use Case Descriptions
→
LLM (Verification)
→
Requirement Analyst (Feedback)
→
Improved Use Case Descriptions
The UCGen approach streamlines use case generation and verification through a human-in-the-loop pipeline. It starts with problem specification, moves to LLM-powered generation, then verification with analyst feedback, leading to improved descriptions.
97.52%
Correctness Post-HITL
| Feature | LLM-Generated | Human-Derived |
|---|---|---|
| Completeness | Higher (avg. 90%+, often 100%) | Lower (sometimes <60%) |
| Correctness | Higher (avg. 92.93% pre-HITL, 97.52% post-HITL) | Lower |
| Redundancy | Lower (mostly <3%) | Higher (large overlaps) |
| Consistency | Good (due to rule-aware prompting) | Variable |
| Usability/Understandability | Improved | Variable |
Impact of Human-in-the-Loop (HITL)
The study found that integrating a human-in-the-loop (HITL) mechanism significantly improved the quality of LLM-generated use case descriptions. By actively seeking human responses to clarification questions, the system could resolve natural language ambiguities, leading to enhanced correctness and completeness. This iterative refinement process, however, sometimes introduced a slight increase in redundancy due to the expansion of use case scenarios and detailed information in pre/post-conditions.
0
Correctness Increase Post-HITL
Advanced ROI Calculator
Estimate the potential return on investment for integrating UCGen-like AI solutions into your enterprise requirements engineering processes.
Estimated Annual Savings
$0
Annual Hours Reclaimed
0
Implementation Roadmap
A phased approach to integrate UCGen capabilities into your existing software development lifecycle for maximum impact.
Automated Generation (Current)
LLM-powered pipeline for goal identification, actor/use case extraction, and scenario generation using tailored prompting strategies.
Human-in-the-Loop Verification (Current)
Analyst feedback through clarification questionnaires for iterative refinement and quality improvement.
SLM Adaptation (Future)
Explore parameter-efficient tuning for Small Language Models (SLMs) to enhance accessibility and reduce computational costs.
Custom Dataset Creation (Future)
Develop a specialized dataset of problem specifications with annotated use case descriptions for fine-tuning LLMs.
Ready to Transform Your Enterprise?
Schedule a personalized consultation to explore how UCGen can revolutionize your requirements engineering process.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat