Vision to Specification: Automating the Transition from Conceptual Features to Functional Requirements

TL;DR

This paper addresses the challenge of transforming high-level, abstract features into concrete, testable functional requirements (FRs). It introduces EasyFR, a three-phase pipeline that first induces SRL-based templates from diverse FRs, then uses a dual-task Key2Temp model to select a template variant and map feature tokens to slots, and finally fine-tunes ChatGLM-6B with P-Tuning v2 to draft FR statements. Across four open datasets and two deployment scenarios, EasyFR frequently outperforms strong baselines (GENIUS, BART) and rivals GPT-4 on key robustness metrics like METEOR and NIST, while leveraging a compact model size. The study demonstrates that SRL-guided, template-constrained generation substantially improves the clarity and consistency of FRs and offers practical pathways for integration into requirements engineering tools, with notable gains when pre-existing FRs are available. Limitations include generation of single-sentence FRs, reliance on annotated SRL roles, and the need for broader real-world validation across more domains and models.

Abstract

The translation of high-level abstract features into clear, and testable functional requirements (FRs) is a crucial step in software development, bridging the gap between user needs and technical specifications. In engineering practice, significant expert effort is needed for this translation. Our approach, EasyFR, streamlines the process by recommending Semantic Role Labeling (SRL) sequences for the given abstract features to guide Pre-trained Language Models (PLMs) in producing cohesive FR statements. By analyzing ten diverse datasets, we induce two variable SRL templates, each including two configurable parts. For concrete features, our proposed Key2Temp model can construct the appropriate variant of the SRL template by identifying a variable SRL template and placing the feature tokens in the appropriate slots. In this way, our approach reframes the process of requirement generation into a structured slot-filling activity. Experimental validation on four open datasets demonstrates that EasyFR outperforms three advanced Natural language generation (NLG) approaches, including GPT4, particularly when existing FRs are available for training. The positive influence of our SRL template variant recommendations is further confirmed through an ablation study. We believe that our results indicate a notable step forward in the realm of automated requirements synthesis, holding potential to improve the process of requirements specification in future software projects.

Figures (4)

• Figure 1: An illustrative example showing our target and design.
• Figure 2: Overview procedure of our approach EasyFR.
• Figure 3: The training architecture of Key2Temp.
• Figure 4: The prompt used for GPT-4-0613 in this study.