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	2024/07/16
形態種別	国際会議プロシーディングス
査読	査読あり
標題	Analysis of Quality Improving of Source Code Generation by LLM along with More Detailing of Development Documents
執筆形態	共著
掲載誌名	Proceedings of ICIS 2024-Summer III
掲載区分	国外
出版社・発行元	ACIS International
総ページ数	6
担当区分	最終著者
著者・共著者	Reon NOMOTO, Hironori TAKUMA
概要	In this study aims to provide methodologies for generating source code using prompts primarily based on natural language and analyzes the impact on the quality of software constructed using source code generated from detailed development documents. We reflected requirements definition, basic design, and detailed design in three types of prompts incrementally and performed source code generation using Large Language Models. We conducted 39 tests in use-case format. The results showed that as the level of detail in the information reflected in the prompts increased, a tendency for improved software quality was seen. However, it was also noted that there is a need for users to explicitly include exception handling in the prompts, and with the increase in information and input token count, there is a higher possibility that some implementations may be omitted.