|
タクマ ヒロノリ
Hironori Takuma
田隈 広紀 所属 千葉工業大学 未来変革科学部 デジタル変革科学科 千葉工業大学 社会システム科学研究科 マネジメント工学専攻 職種 教授 |
|
| 言語種別 | 英語 |
| 発行・発表の年月 | 2025/06/17 |
| 形態種別 | 学術雑誌 |
| 査読 | 査読あり |
| 標題 | Investigating the Relationship Between Quality and Prompt Specificity in Source Code Generation by LLM |
| 執筆形態 | 共著 |
| 掲載誌名 | Computer and Information Science and Engineering |
| 掲載区分 | 国外 |
| 出版社・発行元 | Springer Nature |
| 巻・号・頁 | (1192),pp.101-113 |
| 総ページ数 | 14 |
| 担当区分 | 最終著者 |
| 著者・共著者 | Leon Nomoto, Hironori Takuma |
| 概要 | In this study aims to provide a methodology for generating source code based on natural language prompts and analyze the impact of detailed development documentation on the quality of software constructed using the generated source code. Using three types of prompts that progressively reflect a requirements specification, a basic design document, and a detailed design document, source code was generated by an LLM, and 39 tests were conducted in the form of use cases. As a result, a trend was observed in which the quality of the software improved as the level of detail reflected in the prompt increased. Furthermore, it was found that for very small-scale software, most functions could be implemented by presenting only the functional requirements. However, the need for users to explicitly specify items that do not pertain to functional compatibility, such as exception handling, and the necessity of measures to address the increase in input token count were pointed out. |
| DOI | https://doi.org/10.1007/978-3-031-82606-1_7 |
| ISBN | 978-3-031-82606-1 |