One of the core practices of science is constructing scientific explanations. However, numerous studies have shown that constructing scientific explanations poses significant challenges to students. Proper assessment of scientific explanations is costly and time-consuming, and teachers often do not have a clear definition of the educational goals for formulating scientific explanations. Consequently, teachers struggle to support their students in this process. It is hoped that recent advances in machine learning (ML) and its application to educational technologies can assist teachers and learners in analyzing student responses and providing automated formative feedback according to well-defined pedagogical criteria. In this study, we present a method to automate the entire assessment-feedback process. First, we developed a causal-mechanical (CM)-based grading rubric and applied it to student responses to two open-ended items. Second, we used unsupervised ML tools to identify patterns in student responses. Those patterns enable the definition of “meta-categories” of explanation types and the design of personalized feedback adapted to each category. Third, we designed an in-class intervention with personalized formative feedback that matches the response patterns. We used natural language processing and ML algorithms to assess students' explanations and provide feedback. Findings from a controlled experiment demonstrated that a CM-based grading scheme can be used to identify meaningful patterns and inform the design of formative feedback that promotes student ability to construct explanations in biology. We discuss possible implications for automated assessment and personalized teaching and learning of scientific writing in K-12 science education.

中文翻译：

---

生物学中的因果机械解释：在科学课堂上应用自动评估进行个性化学习

科学的核心实践之一是构建科学解释。然而，大量研究表明，构建科学解释给学生带来了重大挑战。对科学解释的正确评估既昂贵又耗时，而且教师往往对制定科学解释的教育目标没有明确的定义。因此，教师在此过程中努力支持学生。希望机器学习 (ML) 及其在教育技术中的应用的最新进展能够帮助教师和学习者分析学生的反应，并根据明确的教学标准提供自动化的形成性反馈。在这项研究中，我们提出了一种自动化整个评估反馈过程的方法。首先，我们开发了一个基于因果机械 (CM) 的评分标准，并将其应用于学生对两个开放式项目的回答。其次，我们使用无监督的机器学习工具来识别学生反应的模式。这些模式可以定义解释类型的“元类别”，并设计适合每个类别的个性化反馈。第三，我们设计了一种课堂干预，其中包含与反应模式相匹配的个性化形成性反馈。我们使用自然语言处理和机器学习算法来评估学生的解释并提供反馈。对照实验的结果表明，基于 CM 的评分方案可用于识别有意义的模式，并为形成性反馈的设计提供信息，从而提高学生构建生物学解释的能力。我们讨论了 K-12 科学教育中科学写作的自动化评估和个性化教学的可能影响。