Research on student modelling and problem generation for intelligent CAI systems

该项目旨在实现高度个性化的对话。为了达到这一目的，我们研究了以下四个主题：1）智能CAI的概念模型：为了使混合主动性和高度个性化成为可能，ICAI系统不仅要认识到教学的目的和解决意外问题，而且要了解学生的需求，学生解决问题的方法和对话上下文。为了满足上述需求，我们提出了一个对话系统的概念模型，称为“对偶模型”。2）知识表示模型：我们提出了一个多层次的知识表示模型来实现对偶模型。在该模型中，提供了两种知识继承关系：自然蕴涵关系和策略图。策略图将所有课件知识按照世界的教学顺序划分为小世界。策略图的引入产生了以下两个特征。首先，学生可以很容易地理解的材料，通过研究它一步一步沿着战略图。第二，策略图有助于减少系统识别学生错误源时的搜索范围。3）学生模型：提出了一种“扰动法”来识别学生错误源。许多学生的错误被认为是源于常见的行为，这是独立于课件材料，这些行为包括记住错误的事实，错误地应用一些解决问题的方法，混合新学习的知识与以前学习的，等等。相应于这些错误的行为，我们的方法产生的假设，对学生的错误知识扰动正确的知识。4）教学对话：研究了问题生成、教学补救和学生问题回答的方法。

This project aims to realize highly individualized dialogue. In order to achieve the purpose, we studied following four themes.1) Conceptual model of Intelligent CAI:In order to make mixed initiative and highly individualization feasible, it is necessary for an ICAI system not only to recognize the purpose of tutoring and to solve unexpected problems but also to understand student's requirements, student's method of problem solving and dialogue context. In order to fulfill the above requirements, we proposed a conceptual model of a dialogue system called "dual model".2) Knowledge representation model: We proposed a multi-hierarchical knowledge representation model to implement the dual model. In this model, two kinds of relationship for knowledge inheritance are provided; natural implication relations and strategy graph. The strategy graph divides all courseware knowledge into small worlds according to the tutoring sequence of worlds. The introduction of the strategy graph results the following two features. First, students can easily understand the material by studying it step by step along the strategy graph. Second, the strategy graph contributes to reduce the search spece when the system identifies student's error origins.3) Student model: We proposed a method called " perturbation method" to identify student's error origins. Many of students' error are considered to originate from commonly observed behaviors which are independent from courseware materials; those behaviors include remembering wrong facts, applying some methods of problem solving incorrectly, mixing newly studied knowledge with formerly studied one, etc. Corresponding to these erroneous behaviors, our method generates hypotheses against student's erroneous knowledge by perturbing correct knowledge. The identified error origins are accumulated into each student's model.4) Pedagogical dialogue: We studied on methods of generating questions,pedagogical remediation and answering student's questions.