本研究旨在探討如何透過生成式AI技術產生程式設計教育中的個人化程式行為回饋建議，使學生在課堂中時能夠比一般複習活動更能提高學習成效；同時探討透過生成式AI所產生的程式練習題是否可以達到人工出題的品質。 為了實現符合學生的個人化程式回饋建議，此回饋基於學生編碼習慣去識別出的Coding pattern來產生相對應的回饋，並在每次課堂結束後給予學生個人化回饋以及考前的干預輔導活動，結果顯示有個人化干預回饋的幫助，學生學習成效確實有效提升。 為了評估題目的可信度，將透過生成式AI產生的程式練習題與人工題目混合並請專業人士進行評估，將評估結果進行Fleiss Kappa，而透過Kappa係數去驗證其題目的信度與一致性。同時在課堂中將生成式AI產生的題目與人工題目混雜著給學生在課堂上做練習題，並在每週課堂結束後，請學生透過自己在做題時的感覺猜測哪些題目為生成式AI出題哪些為人工出題，圖靈測試結果表明生成式AI產生的題目與人工題目在學生感受下是無法分辨的。 以上研究皆表明，生成式AI的出現可以帶來許多的便利性及可能性，除了減輕老師的壓力，也能夠確實幫助學生提高學習成效，未來許多的應用必定會使用到大量的生成式AI。;This study aims to explore how Generative AI technology can enhance personalized programming behavior feedback in programming education, enabling students to achieve better learning outcomes during classes compared to traditional review activities. Additionally, the study investigates whether programming exercises generated by Generative AI can achieve the same quality as those created by humans. To provide personalized programming feedback, this feedback is based on identifying coding patterns from students′ coding habits and generating corresponding feedback, which is given to students at the end of each class. The results show that the feedback helps improve students’ learning effectiveness. To evaluate the reliability of the generated exercises, programming exercises produced by Generative AI will be assessed by professionals alongside manually created questions. The reliability and consistency of the questions will be validated using the Kappa coefficient of Fleiss′ Kappa. During classes, both AI-generated and human-created questions will be mixed for students to solve. At the end of each weekly class, students will be asked to guess which questions were generated by AI and which were created by humans. The Turing test results indicate that it is impossible to distinguish between AI-generated and human-created questions. The above findings demonstrate that the advent of Generative AI brings many conveniences and possibilities. In addition to reducing teachers′ workload, it effectively helps students improve their learning outcomes. In the future, many applications will undoubtedly utilize extensive Generative AI.
