Scholarlink

<p>Software requirements are the driving force behind software development. As the cornerstone of the entire software lifecycle, the efficiency of crafting requirement specifications and the quality of these requirements significantly influence the duration of software development. Despite massive research on requirements elicitation, the reality is that requirements are often painstakingly crafted manually, word by word. This manual process is not only time-consuming but also prone to issues such as the misuse of terminology. To address these challenges, we introduce <i>ReqCompletion</i>, an approach designed to recommend the next token in real-time for given prefix of requirements description. <i>ReqCompletion</i> comprises two phases.Firstly, we have designed and implemented the token completion approach. We integrate a knowledge-injection module and a pointer network into GPT-2, which stands as the largest available GPT model that allows for fine-tuning on specialized downstream tasks. Second, we design a distil phase to speed up the completion by downsizing the <i>ReqCompletion</i>. Empirical evaluations using two public datasets demonstrate that <i>ReqCompletion</i> surpasses all baselines in performance (Recall@7 gains up to 65.87% than the second-best model). Furthermore, the effectiveness of its two pivotal design elements has been substantiated through rigorous ablation studies. Regarding the distillation phase, we downsized the model to just 11.6% of its original size, resulting in a 4.07-fold speedup while maintaining a recall value of 98.9%. The utility of our work has been evaluated preliminarily through a small user study.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

ReqCompletion: domain-enhanced automatic completion for software requirements

  • 期刊论文

  • 发表日期：

    2025年4月17日

查看全文
  • Xiaoli Lian,
  • Jieping Ma,
  • Heyang Lv,
  • Li Zhang

摘要

Software requirements are the driving force behind software development. As the cornerstone of the entire software lifecycle, the efficiency of crafting requirement specifications and the quality of these requirements significantly influence the duration of software development. Despite massive research on requirements elicitation, the reality is that requirements are often painstakingly crafted manually, word by word. This manual process is not only time-consuming but also prone to issues such as the misuse of terminology. To address these challenges, we introduce ReqCompletion, an approach designed to recommend the next token in real-time for given prefix of requirements description. ReqCompletion comprises two phases.Firstly, we have designed and implemented the token completion approach. We integrate a knowledge-injection module and a pointer network into GPT-2, which stands as the largest available GPT model that allows for fine-tuning on specialized downstream tasks. Second, we design a distil phase to speed up the completion by downsizing the ReqCompletion. Empirical evaluations using two public datasets demonstrate that ReqCompletion surpasses all baselines in performance (Recall@7 gains up to 65.87% than the second-best model). Furthermore, the effectiveness of its two pivotal design elements has been substantiated through rigorous ablation studies. Regarding the distillation phase, we downsized the model to just 11.6% of its original size, resulting in a 4.07-fold speedup while maintaining a recall value of 98.9%. The utility of our work has been evaluated preliminarily through a small user study.