Tailoring Personalized Mathematics Education for High School Students Using Dynamic Treatment Regimes

美国大部分地区目前针对高中生的数学选课计划并没有针对每个学生的数学成绩进行优化，尤其是在高等数学方面。特别是，不同种族/族裔和收入群体参加高等数学课程的比例不成比例。为此，这项研究项目利用个性化医学的最新进展，开发了数据驱动的、个性化的数学课程计划。在个性化医疗中，临床医生在机器学习的帮助下，使用过去和当前的患者历史数据来定制针对患者的治疗计划。同样，拟议的数据驱动的个性化选课计划通过使用当前和过去学生表现的大规模教育数据，在适当的时间为学生推荐合适的数学课程。此外，该项目纳入了来自计算机科学和统计的算法公平约束，以确保建议减少种族/族裔和收入群体之间现有的选课模式差异。更广泛地说，该项目希望通过拟议的数据驱动的个性化建议，改变学生学习数学课程(或更广泛地说，K-12 STEM课程)的方式，使每个学生，特别是来自代表性不足或弱势背景的学生，都有机会攻读STEM专业和从事职业。为了制定个性化的数学选课计划，该项目使用了一套被称为最佳动态治疗方案(OTRS)的统计技术。使用OTRS的关键一步是从收集的数据中了解治疗效果的异质性。因此，该项目的第一个目标是利用现有的使用机器学习的工作来评估在大规模教育研究中学习不同数学课程的不同影响。接下来，该项目使用Q学习、A学习和价值搜索方法来开发OTRs，特别是数学选课的顺序决策。最后，为了减轻拟议OTRs中与公平相关的危害，并确保OTRS不会产生歧视性建议，该项目利用了算法公平文献中的公平约束。该项目将把当前最先进的OTR方法转化为K-12数学教育政策，并将首次尝试将算法公平性和最佳政策学习结合起来，为STEM教育设计公平的政策。该项目由NSF的STEM教育研究中的EHR核心研究能力建设计划(ECR：BCSER)支持，该计划旨在培养研究人员开展高质量STEM教育研究的能力。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。

The current math course-taking plan for high school students across much of the United States is not optimized for every student’s success in math, especially in advanced mathematics. In particular, there has been disproportional participation in advanced math courses among different racial/ethnic and income groups. To this end, this research project develops data-driven, personalized math course-taking plans by leveraging recent advances in personalized medicine. In personalized medicine, clinicians, with the help of machine learning, use past and current historical patient data to tailor patient-specific treatment plans. In a similar vein, the proposed data-driven, personalized course-taking plan recommends appropriate math courses for students at the opportune time by using large-scale educational data from current and past students’ performances. Additionally, the project incorporates algorithmic fairness constraints from computer science and statistics to ensure that the recommendations reduce existing disparities in course-taking patterns among racial/ethnic and income groups. More broadly, with the proposed, data-driven personalized recommendations, the project hopes to transform how students take math courses (or more broadly, K-12 STEM courses) so that every student, especially students from underrepresented or disadvantaged backgrounds, will opportunities to pursue STEM majors and careers. To develop personalized math course-taking plans, the project uses a set of statistical techniques known as optimal dynamic treatment regimes (OTRs). A crucial step in using OTRs is understanding treatment effect heterogeneity from the collected data. Consequently, the first aim of the project is to leverage the existing work on using machine learning to estimate heterogeneous effects of taking different math courses in large-scale educational studies. Next, the project uses Q-learning, A-learning, and value search methods to develop OTRs, specifically sequential decisions for math course-taking. Finally, to mitigate fairness-related harms in the proposed OTRs and to ensure that the OTRs do not produce discriminatory recommendations, the project utilizes fairness constraints from the algorithmic fairness literature. The project will translate the current state-of-the-art OTR methods into K- 12 math education policies and will constitute the first attempt to integrate algorithmic fairness and optimal policy learning to design equitable policies for STEM education. The project is supported by NSF's EHR Core Research Building Capacity in STEM Education Research (ECR: BCSER) program, which is designed to build investigator's capacity to carry out high-quality STEM education research.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.