Eliciting Adaptive Sequences for Online Learning

适应性学习旨在为每个学生提供高效、有效和定制的学习路径。适应性学习系统的目的不仅在于熟练掌握和确定学生真正知道什么，而且还在于通过顺序学习路径准确、有逻辑地引导学生达到规定的学习成果和技能掌握。这些自适应学习系统正在迅速涌现，但仍处于实验阶段。尽管它们有很多好处——比如提供更高的时间效率和集中的补救措施——但开发它们是一项挑战。自适应排序的内容和评估是自适应学习系统的核心。现有的自适应测序方法，如基于规则的方法、基于部分观察马尔可夫决策过程（POMDP）的框架，要么是无效的，要么是难以开发的。学习分析、数据挖掘和机器学习（特别是多臂强盗算法）的最新进展为构建用于在线学习的有效自适应排序算法提供了新的机会。多臂强盗算法家族以一个赌徒命名，他必须在一系列试验中决定拉动“多臂强盗”老虎机的哪只手臂以最大化总奖励。这些算法是数据驱动的，可以平衡探索和开发，并在不确定的情况下做出顺序决策。因此，它们与替代教学法的决策特别相关，并且很自然地有助于确定在线学习环境中自适应序列的问题。本研究的目标是：(1)设计一种系统的方法，其中包括一套基于多臂框架的算法，该算法将与分层教学模型和学生模型一起工作，以在自适应学习系统中产生知识成分、学习活动和形成性评估问题的自适应序列。(2)使用学生和课程的模拟来探索设计算法的性能，并优化指标或奖励策略；(3)开发一个自适应学习系统的原型，以便在真实的在线课程中对设计的算法的性能进行评估。本研究的长期目标是使未来的在线学习系统能够实时地提供自适应地改变学习序列的内容和活动，这将最适合学生的需求和知识状态。这样的系统不仅会使学生的学习更容易，而且会大大提高效率。

Adaptive learning aims to provide efficient, effective, and customized learning paths to engage each student. The intent of adaptive learning systems is not only to use proficiency and determine what a student really knows but also to move students accurately and logically through a sequential learning path to attain the prescribed learning outcomes and skill mastery. These adaptive learning systems are quickly emerging but are still in the experimental stages. Although they have many benefits - such as providing greater time efficiency and focused remediation - it is challenging to develop them. Adaptively sequencing content and assessment is the core of an adaptive learning system. Existing adaptive sequencing approaches, such as the rule-based approach, the partially observed Markov decision process (POMDP) based framework, are either ineffective or difficult to develop. Recent advances in learning analytics, data mining, and machine learning (multi-armed bandit algorithms in particular), present new opportunities for building effective adaptive sequencing algorithms for online learning. The multi-armed bandit family of algorithms is named after a gambler who must decide which arm of a "multi-armed bandit" slot machine to pull to maximize the total reward in a series of trials. These algorithms are data-driven, can balance exploration and exploitation, and make sequential decisions under uncertainty. Thus, they are particularly relevant to decision-making about alternative pedagogies and lend themselves quite naturally to the problem of determining adaptive sequences in online learning environments. The objectives of this research are (1) to design a systematic methodology that includes a suite of algorithms based on the multi-armed framework that will work together with a hierarchical pedagogical model and a student model to produce adaptive sequences of knowledge components, learning activities, and formative assessment questions in an adaptive learning system. (2) to use simulations of students and courses to explore the performance of the designed algorithms and to optimize metrics or reward strategies; and (3) to develop a prototype of adaptive learning systems that will enable evaluations of the performance of the designed algorithms in real online courses. The long-term goal of this research is to enable future online learning systems to provide adaptively altering learning sequences of content and activities in real time that will best fit the student's needs and knowledge states. Such systems are expected to make student learning not only easier but also far more efficient.