CAREER: Combining Human Judgment and Data-Driven Approaches for the Development of Interpretable Models of Student Behaviors: Applications to Computer Science Education

调试，即识别和解决计算机程序中的缺陷的过程，是学习编程时需要掌握的一项重要技能。然而，许多对编程有很好理解的新手程序员在调试过程中举步维艰。尽管如此，调试很少被明确教授。这个项目将使用数据驱动的方法来研究注册在大学水平的计算机科学入门课程的新手程序员的调试过程，确定他们调试行为的有意义的元素，以及这些元素如何结合在一起形成共同的调试策略。这些知识将被用来创建能够识别学生在编程时使用哪些调试策略的计算机算法。这些算法将设计为学生和教师可以解释。这些算法将用于支持学生学习有效的调试策略，帮助教师监控学生的调试实践，并帮助未来的K-12教师学习调试过程中有意义的元素。这个项目的结果将通过支持未来的教师学习如何培养重要的调试技能，对大学计算机科学入门课程和K-12水平的计算机科学教育的状况产生积极的影响。该项目将通过将知识工程和机器学习相结合的方法来创建可解释的学生行为模型，从而促进学生行为建模的最新水平。它将提供经验证据，说明学生行为模型的可解释性如何提供强大的教学优势，而不仅仅是它在预测学生行为方面的准确性。所提出的方法将被应用于大学计算机科学入门课程中的调试策略研究，通过从在线问题解决平台PrairieLearn收集的日志数据。可解释模型的好处将与传统机器学习方法的好处进行比较，使用严格的研究来确定支持学生和教师的最佳方法。具体地说，该项目将应用这种方法，利用其创建的模型的更高的可解释性，以：(1)更好地了解学生的调试行为；(2)支持学生对其调试策略进行自我反思并发展有效的调试实践；(3)为教师提供有关学生调试过程的可操作信息；以及(4)支持未来的教师获得关于学生调试策略的假设的专业知识。通过这样做，该项目将有助于新手程序员对调试过程的一般知识，并建立方法来支持大学生和未来的K-12教师获得关于调试过程的明确知识。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Debugging, the process of identifying and resolving defects in computer programs, is an important skill to acquire while learning to program. However, many novice programmers, with good understanding of programming, struggle with the debugging process. Despite this, debugging is rarely explicitly taught. This project will use data-driven approaches to study the debugging processes of novice programmers enrolled in a college level introductory computer science course, identify the meaningful elements of their debugging behaviors and how those elements combine to form common debugging strategies. This knowledge will be used to create computer algorithms that are able to identify which debugging strategies students use when programming. These algorithms will be designed to be interpretable by students and instructors. These algorithms will be used to support students in learning efficient debugging strategies, assist instructors in monitoring their students' debugging practices, and help future K-12 teachers in learning about the meaningful elements of the debugging process. The results of this project will positively impact the state of computer science education in both college level introductory computer science courses and in the K-12 level by supporting future teachers in learning how to foster important debugging skills.The project will contribute to the state of the art in student behavior modeling by formalizing an approach that combines knowledge engineering and machine learning to create interpretable models of student behavior. It will provide empirical evidence illustrating how the interpretability of a student behavior model can provide powerful pedagogical advantages beyond its accuracy at predicting a student's behavior. The proposed approach will be applied to study debugging strategies in college level introductory computer science courses through the log data collected from an online problem-solving platform named PrairieLearn. The benefits of interpretable models will be compared to those of traditional machine learning approaches, using rigorous research to identify the best methods for supporting students and instructors. Specifically, the project will apply this approach, leveraging the increased interpretability of the models it creates, to: (1) better understand students' debugging behaviors; (2) support students in self-reflecting about their debugging strategies and developing efficient debugging practices; (3) provide instructors with actionable information about their students' debugging processes; and (4) support future teachers in acquiring expertise in formulating hypotheses about students' debugging strategies. By doing so, the project will contribute to general knowledge about debugging processes for novice programmers, and establish methods to support college students and future K-12 teachers in acquiring explicit knowledge about the debugging process.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.