Modeling Interactions Across Skills: A Method to Construct and Compare Models Predicting the Existence of Skill Relationships
Modeling Interactions Across Skills: A Method to Construct and Compare Models Predicting the Existence of Skill Relationships
复制标题
跨技能交互建模：构建和比较预测技能关系存在的模型的方法
DOI：
--
复制
发表时间：
2016
期刊：
影响因子：
--
通讯作者：
N. Heffernan
中科院分区：
文献类型：
--
作者：
Anthony F. Botelho;Seth A. Adjei;N. Heffernan
将先决技能结构纳入教育系统有助于确定概念应该呈现给学生的顺序，以优化学生的成绩。许多技能具有因果关系，其中一项技能必须先于另一项技能呈现，这表明一种强大的技能关系。了解这种关系可以帮助预测学生的表现并确定先决条件。然而，技能关系并不能直接测量；相反，这种关系可以通过观察学生在不同技能上的表现差异来估计。然而，这些估计方法似乎缺乏一个基线模型来比较它们的有效性。如果估算关系存在性的两种方法产生两个不同的值，哪一种结果更准确？在这项工作中，我们提出了一种比较模型的方法，试图衡量技能关系的强度。通过这种方法，我们开始识别那些学生水平的协变量，这些协变量提供了预测技能关系存在的最准确模型。着眼于跨技能绩效的相互作用，我们使用我们的方法构建模型来预测五种强相关技能对和五种模拟弱相关技能对的存在。我们的方法能够评估几个模型，这些模型区分了这些差异，并在零模型上获得了显著的准确性增益，并提供了一种方法来识别学生掌握的互动对我们分析中的这些增益提供了最重要的贡献。
The incorporation of prerequisite skill structures into educational systems helps to identify the order in which concepts should be presented to students to optimize student achievement. Many skills have a causal relationship in which one skill must be presented before another, indicating a strong skill relationship. Knowing this relationship can help to predict student performance and identify prerequisite arches. Skill relationships, however, are not directly measurable; instead, the relationship can be estimated by observing differences of student performance across skills. Such methods of estimation, however, seem to lack a baseline model to compare their effectiveness. If two methods of estimating the existence of a relationship yield two different values, which is the more accurate result? In this work, we propose a method of comparing models that attempt to measure the strength of skill relationships. With this method, we begin to identify those student-level covariates that provide the most accurate models predicting the existence of skill relationships. Focusing on interactions of performance across skills, we use our method to construct models to predict the existence of five strongly-related and five simulated poorly-related skill pairs. Our method is able to evaluate several models that distinguish these differences with significant accuracy gains over a null model, and provides the means to identify that interactions of student mastery provide the most significant contributions to these gains in our analysis.