- 學位論文
Chatbot提示明確性對網頁初學者開發網站原型的效用評估-以Chatgpt協作開發SSDLC文件化管理系統為例
Evaluating the Effectiveness of Chatbot Prompt Clarity on Website Prototype Development by Novice Web Developers: A Case Study of Chatgpt Collaborative Development of SSDLC Documentation Management System
摘要
自從OpenAI正式發布開發的人工智慧聊天機器人-ChatGPT(Chat Generative Pre-trained Transformer)後，不僅帶來AI劃時代的變革，更掀起全世界研究ChatGPT應用範圍的風潮，從日常的聊天、創作文章、寫履歷、寫程式甚至考檢定考試都有優異的表現。本論文以初學網頁開發者的角色運用免費版ChatGPT生成程式碼功能開發系統為研究主軸。首先以因應資通安全責任等級分級辦法附表十資安法防護基準表中「系統與服務獲得」構面要求，開發安全軟體開發生命週期文件化管理系統 (SSDLC Documentation Management System)以留存各開發階段產出的相關文件為動機，再以ChatGPT擔任虛擬開發工程師的角色，運用提示工程(Prompt Engineering)的文本提示技術生成網頁前後端代碼，以敏捷式迭代(iteration)的方式開發系統，記錄整個溝通過程以提出人工智慧聊天機器人生成程式碼的發現事項。最後基於開發過程中與ChatGPT溝通的經驗，進一步實驗以五種不同明確性的提示情境運用免費版ChatGPT開發七種常見網頁功能模組，用計算對話token數的量化方式評估提示明確性高低是否影響ChatGPT生成代碼的效用，並依據實驗結果提出有效的提示範本供開發團隊中的專案經理或產品經理運用，以快速生成可運作的網頁原型與客戶確認需求功能，藉以縮短軟體開發時間。
並列摘要
Since the official release of ChatGPT (Chat Generative Pre-trained Transformer) developed by OpenAI, it has not only brought about a revolutionary change in AI but also sparked a global trend in researching the application scope of ChatGPT. It has demonstrated outstanding performance in various tasks such as daily conversation, content creation, resume writing, programming, and even exam preparation. This thesis focuses on how the role of novice web developers can leverage the free version of ChatGPT to generate code for web development. The motivation for this research arises from the need of web development of a Secure Software Development Lifecycle Documentation Management System (SSDLC Documentation Management System) is initiated to retain relevant documents produced at each development stage. ChatGPT assumes the role of a virtual developer, utilizing the Textual Prompting technique within the framework of Prompt Engineering to generate both front-end and back-end code for web development. The system is developed in an agile iterative manner, documenting the entire communication process to present insights into the use of AI chatbots for code generation. Based on experiences gained from communication with ChatGPT during the development process, further experiments are conducted. Using five different scenarios with varying levels of precise prompts, the free version of ChatGPT is employed to develop seven common web functionality modules. The utility of ChatGPT-generated code is quantitatively evaluated by calculating the number of conversation tokens. The research aims to determine whether prompt preciseness affects the effectiveness of ChatGPT in code generation. Based on the experimental results, effective prompt templates are proposed for use by project managers or product managers in development teams. These templates aim to facilitate the rapid generation of functional web prototypes and confirmation of customer requirements, ultimately reducing software development time.