FAST: Feature-Aware Student Knowledge Tracing
FAST: Feature-Aware Student Knowledge Tracing
复制标题
快速：具有特征意识的学生知识追踪
DOI：
--
复制
发表时间：
2013
期刊：
影响因子：
--
通讯作者：
Peter Brusilovsky
中科院分区：
文献类型：
--
作者：
José P. González;Yun Huang;Peter Brusilovsky
各种电子学习系统，如大规模开放在线课程和智能辅导系统，现在正在从学生随着时间的推移解决不同熟练程度的问题产生大量功能丰富的数据。为了分析这些数据，研究人员经常使用知识追踪[4]，这是一种有20年历史的方法，已经成为从成绩数据推断学生知识的事实上的标准。知识跟踪使用隐马尔可夫模型(HMM)从学生的成绩答题中估计潜在的认知状态(学生的知识)。由于原始的知识跟踪公式不允许对一般特征建模，因此相当多的研究集中在对知识跟踪算法的特别修改上，以使能够对感兴趣的特定特征进行建模。这导致了太多不同的知识，为了非常具体的目的而重新制定。例如，Pardos等人。[5]Beck等人提出了一个新的模型来衡量学生个体特征的影响。[2]修改了知识追踪来评估辅导系统中帮助的效果，Xu和Mostow[7]提出了一个新的模型，允许测量子技能的效果。这些特别模型成功地实现了它们各自的特定目的，但它们并不适用于任意功能。已经提出了允许更灵活的特征的其他学生建模方法。例如，性能因素分析[6]使用逻辑回归对任意特征进行建模，但不幸的是，它不能推断学生是否已经学习了一项技能。提出了一种将一般特征引入到知识跟踪中的新方法FAST(Feature-Aware Study Knowledge Tracing)。FAST将性能因素分析(Logistic回归)与知识跟踪结合在一起，利用了以前关于无监督学习的工作[3]。因此，FAST能够像知识追踪那样推断学生的知识，同时也允许像性能因素分析那样的任意特征。FAST允许将一般特征用于知识跟踪，方法是用Logistic回归[3]取代生成发射概率(通常称为猜测和失误概率)，因此这些概率可以随时间变化来推断学生的知识。FAST允许任意特征联合训练Logistic回归模型和HMM。同时训练参数使FAST能够从功能中学习。这不同于使用回归来分析失误和猜测概率[1]。为了验证我们的方法，我们使用了从与导师互动的真实学生那里收集的数据。我们给出了与知识跟踪和性能因子分析相比较的实验结果。根据绩效因素分析的公式，我们使用我们的模型进行了实验，使用了项目难度、学生与项目相关的技能(或多项技能)的先前成功和失败等特征。
Various kinds of e-learning systems, such as Massively Open Online Courses and intelligent tutoring systems, are now producing amounts of feature-rich data from students solving items at different levels of proficiency over time. To analyze such data, researchers often use Knowledge Tracing [4], a 20-year old method that has become the de-facto standard for inferring student’s knowledge from performance data. Knowledge Tracing uses Hidden Markov Models (HMM) to estimate the latent cognitive state (student’s knowledge) from the student’s performance answering items. Since the original Knowledge Tracing formulation does not allow to model general features, a considerable amount of research has focused on ad-hoc modifications to the Knowledge Tracing algorithm to enable modeling a specific feature of interest. This has led to a plethora of different Knowledge Tracing reformulations for very specific purposes. For example, Pardos et al. [5] proposed a new model to measure the effect of students’ individual characteristics, Beck et al. [2] modified Knowledge Tracing to assess the effect of help in a tutor system, and Xu and Mostow [7] proposed a new model that allows measuring the effect of subskills. These ad hoc models are successful for their own specific purpose, but they do not generalize to arbitrary features. Other student modeling methods which allow more flexible features have been proposed. For example, Performance Factor Analysis [6] uses logistic regression to model arbitrary features, but unfortunately it does not make inferences of whether the student has learned a skill. We present FAST (Feature-Aware Student knowledge Tracing), a novel method that allows general features into Knowledge Tracing. FAST combines Performance Factor Analysis (logistic regression) with Knowledge Tracing, by leveraging on previous work on unsupervised learning with features [3]. Therefore, FAST is able to infer student’s knowledge, like Knowledge Tracing does, while also allowing for arbitrary features, like Performance Factor Analysis does. FAST allows general features into Knowledge Tracing by replacing the generative emission probabilities (often called guess and slip probabilities) with logistic regression [3], so that these probabilities can change with time to infer student’s knowledge. FAST allows arbitrary features to train the logistic regression model and the HMM jointly. Training the parameters simultaneously enables FAST to learn from the features. This differs from using regression to analyze the slip and guess probabilities [1]. To validate our approach, we use data collected from real students interacting with a tutor. We present experimental results comparing FAST with Knowledge Tracing and Performance Factor Analysis. We conduct experiments with our model using features like item difficulty, prior successes and failures of a student for the skill (or multiple skills) associated with the item, according to the formulation of Performance Factor Analysis.