【作者】 陈乙雄；

【导师】 吴中福；

【作者基本信息】 重庆大学 ， 计算机软件与理论， 2011， 博士

【摘要】 与传统“一对多”的课堂教育相比,远程教育的最大优势在于能够借助各种计算机技术满足学习者的个性化需要,实现“以学生为中心”的差异化教学。为此,近年来一系列支撑个性化学习的应用技术,如智能答疑、自适应测试、个性化推荐等,得到了普遍的发展。然而,通过多年的教学与实践观察,笔者发现:即使教学系统能根据学习者的学习偏好、已掌握知识的情况、甚至是考虑情景因素提供或推荐个性化的学习内容,也不能完全保证学习者能够获得预期的学习效果。这是因为目前多数应用系统都是基于内容而不是基于活动来设计和实现的,这会导致学习者在获得所需的教学内容后缺乏必要的活动引导,使自主学习存在一定的盲目性。从教学规律的角度来看,在相同的教学内容上开展不同的活动,例如一段文本内容可以有朗读、背诵、默写等活动,其获得的学习效果也不相同,因此学习过程应当是由一系列学习活动而非仅是内容组成的。根据上述分析,本文认为当前远程教学系统存在一个迫切需要解决的问题:系统在提供给学习者所需学习资源的同时,如何生成与之对应的学习活动规划与调度方案,从而用以导航服务的方式来指导学习者将学习资源转化为知识和能力。本文的工作就是研究如何采用基于有穷自动机的方法来解决教学支撑平台中学习活动的规划与调度的问题并实现智能导航服务。论文主要从:建立问题的约束满足(Constraint Satisfaction Problem)模型;设计基于有穷自动机的求解方法;在问题规模较大情况下的蚁群优化算法;特定教学模式下的学习活动调度算法等四方面展开,并分别得出相应的结论。1.建立学习活动规划与调度的CSP模型。本文涉及的学习活动规划与调度问题从本质上可看作人工智能领域的智能规划问题,而这类问题常常可以用约束满足问题(即CSP)模型来加以描述。在网络学习环境中,需要将教学目标和教学方法结合起来,在标准化的资源库基础上,首先对学科体系加以规范化;再利用本体技术建立以知识点为核心的课程体系结构;最后根据教学规律定义一系列学习活动(或任务)及这些活动跟知识点的联系。学习者为达到某一学习目标而需要进行的学习活动及其调度方案可以通过CSP模型中的变量来制定,而学习过程中的各约束如访问的并发控制、多个学习者的同步与协同等,可以用约束方程组定义,从而建立起学习活动规划与调度问题的CSP模型。目前,在网络课程的知识点层次上建立学习活动层、并将其转化为约束满足问题来实现智能规划与调度服务的研究尚未见文献报导。2.设计基于有穷自动机的求解方法。在不同的网络学习环境中,约束条件往往可能发生不同的变化,而现有CSP的各类常用求解方法又往往缺乏对约束条件的动态适应操作与创建能力。本文提出的基于自动机的求解方法可以将CSP表示为一个具有最小状态的确定有穷自动机MDFA,该自动机接受的语言即为问题的解。该方法最大的优点在于由环境变化产生的新约束条件可通过简单的运算加入到原有的自动机,使之得到扩展。同时,原先自动机的计算结果也可复用以节省时间开销。在网络学习环境中基于自动机理论对学习活动约束满足问题的算法设计及有关理论研究是本文的创新之处。此外,由于利用自动机理论求解CSP的相关文献很少,因此本文的方法对CSP自身的求解技术发展也具有一定的研究价值。3.规模较大的CSP的蚁群优化算法。蚁群算法的思想源于蚂蚁在寻找食物过程中发现路径的行为方法。由于本文涉及的CSP本质上属于NP问题,在问题规模较大的情况下,可以将CSP转化为一个优化问题求解。有关蚁群优化算法及其应用研究在较多文献已有讨论,本文借鉴了这些文献中蚁群算法的设计方法来实现了解的启发式搜索。4.网络特定教学模式下的学习活动调度算法。在一些特定的网络教学模式下,通过对学习活动CSP模型中的部分变量进行预先赋值,并简化约束条件等方法将问题转化为经典的α|β|γ调度模型求解。目前各类调度模型的求解算法尤其是车间调度模型算法有大量文献可查,但尚无针对网络特定教学模式中的学习活动开展调度模型定义并求解的文章发表。综上所述,在本文涉及的四项主要研究工作中,第一、第四部分的研究成果目前还没有文献报导;第二部分相关文献很少;第三部分虽然已有相关研究成果,但本文着重解决在具体模型中的应用技术问题。总体上本文的创新点主要体现在约束满足问题的建模求解技术及相关的自动机理论研究方面。更多还原

【Abstract】 Compared with the traditional“one to many”educational patterns, the biggest advantage of Elearning lies in its individualized teaching methods which are implemented by various computer technologies. Thus, a series of applications that supports individualized learning have been developed during recent years, such as intelligent FAQ, IRT-based testing, individualized recommendation, etc. However, according to author’s long term observation and teaching experience, the learning effects can not be certified though learning systems can provide learners with invidualized contents which are adapted to their preferences, knowledge level, or even learning circumstance. The cause is that most researches in this field pay too much attention on contents instead of activities, learners are lack of corresponding activity guide after the individualized learning contents are presented. In a pedagogic view, for the same learning material, take a paragraph of text for instance, people can choose to read, recite, or dictate it and obviously the learning effects are different. According to the above analysis, the problem of this thesis can be proposed: how to generate an adaptive activity routine and combine it with individualize learning packges as a guide service so as to assure the final learning effects.This paper aims to solve the proposed study guide problem of learning activities based on CSP model and finite automata. The paper’s work can be presented as the following four parts:1. CSP model of scheduling and planning of learning activities. The propose problem can be converted as AI planning problem and depicted by Constraint Satisfaction Problem (CSP) model. In the context of education, this paper will combine the teaching purpose with its methods, define a discipline framework, set up a standardized teaching resource storage structure, and create a course ontology based on knowledge points. With CSP methods, the purpose, scheduling scheme of a learning process can be defined as variables in CSP model; constraints, such as learning coordination, synchronization, parallel access control, can be defined as constraint equations. Thus, we can establish a CSP scheduling model for e-learning activities.2. A solution based on finite automata. In e-learning environment, constraints are different in various learning scenarios. An automaton based approach shows its advantages in term of constraints handling over other approaches. A CSP problem can be converted as a MDFA whose accepted sentences are solutions of the problem. Using automata to solve CSP problems is rarely discussed so it has theoretical value for research. In addition, the novelty of this paper also lies in the research on applying it to solve learning activities planning and scheduling in e-learning environment.3. Ant colony optimization. The idea of ant colony optimization comes from ants’behavior when seeking food. As the proposed problem is an NP problem, heuristics should be adopted when the problem’s scale is large. If we set the number of conflicts among constraints as the target, the problem can be solved as an optimization problem by ant colony optimization.4. Scheduling algorithms for special teaching scenarios. In some cases, instructors have designed certain teaching scenarios for a well defined learning purpose. Thus, the problem can be simplified as a classic scheduling problem given by a tripleα|β|γ. The way of establishing and solving such scheduling model is discussed in many references especially for job shop scheduling models, but researches focused on categorizing special teaching scenarios in e-learning and setting up corresponded scheduling models are rarely proposed.To summarize, few papers related to the above 1 and 4 researches can be found and the significance of research work 2 and 3 lies in model establishment and its application. This paper is generally an engineering application research.更多还原