导　　师： 叶鲁卿;PIERRE PADILLA

学科专业： G1102

授予学位： 博士

作　　者： ;

机构地区： 华中科技大学

摘　　要： e-education的出现给学习者带来了革命性的变化，用户脚不出户，就可以享受到不受时空限制个性化的远程学习。一般来讲，在e-education环境里，各种资源包括人、电子资源及应用典型地分布在不同的网络节点，这对如何有效地在如此动态、并发、不可预测的分布式环境设计一个自适应、个性化的智能e-education系统带来了巨大的挑战。就目前而言，大多数e-education系统的设计理论来源于目标主义，这种理念来源于传统的教室模型，使得大量的e-books出现在e-education中。从学习者的角度来看，他们难以进行个性化的学习、有效的知识交流和共享和及时对等帮助。从教师的角度来看，他们缺乏一个有效的网上工作环境，比如，在课程内容创作，传授知识，准备试题方面，他们往往不得不进行大量耗时费力的重复工作，尤其是当他们缺乏计算机背景时，更显得不知所从。 为解决上述问题，本文根据最新的教育理论，运用来源于跨学科的多种技术、理论、方法（包括多multi-agent, learning object, genetic algorighm, xml, j2ee等）提出了一个新的e-education体系，其最终目的就是建立一个自适应的、个性化的构建主义学习环境，使得学生在其个人agent的帮助下能够有效地进行个性化的个人或集体学习，同时能够方便教师创作课程内容和组织，准备远程考试。 为了有效地实现这一目标，本文提出了一个基于mas的e-edcuation体系─mage。该体系由众多agent组成，这些agent通过合作、协商和有效的通讯执行不同的子任务，由此形成了一大规模的、复杂的、动态的、开放的、自我组织的多agent,多用户e-education环境。特别地，我们关注下面的建模和应用研究： 在课程创作方面，关键的问题是如何开发能够重用到多种背景下的学习对象并且有效地管理和共享。本文提出了一个新的meeocas体系，包括它的概念模型提出和其中eeo概念的定义以及其内部结构和外部封装的模型。该体系使得课程内容和学习对象创作者能够方便地发现所需的eeo,同时支持eeo定购和发布等系统服务。 在个性化，自适应学习方面，本文提出了一个集成框架，该框架的创新之处在于无缝地集成个人和集体自适应学习。就个人自适应学习空间而言，关键问题是如何根据学生以前的知识状态，学习偏好以及设计的领域模型提供自适应的学习路径和领域概念知识，为此，本文提出了一种基于领域模型的搜索算法；就集体自适应学习空间而言，本文的焦点落在如何设计对等帮助系统和动态地建立学习小组，就此，本文分别提出了相应的体系、模型和策略。在此体系下，由于学习者能够主动选择不同的学习体验，比如，他可以根据自己的需求，通过不同的agent, 要么重新请求新的学习对象，要么请求对等帮助，或者要么自己组织以个学习小组进行有关知识点的学习和讨论。明显地，该系统具有构建主义的特色。 在远程考试方面，传统的系统主要采用client-server模式，这种模式具有显而易见的缺陷，比如，灵活性、伸缩性差同时不能充分支持象自动测试产生、主观题的评估、离线考试，主动告示和考试各环节的通讯等功能。为解决这些问题，本文运用新一代的移动agent理论提出了一个创新的整体解决方案。该方案全方位支持在分布式网络环境下进行试题产生，分发，收集和评估。特别地，本文提出了一个综合运用遗传算法、移动agent和多 agent试题产生体系（gamastp）。在该体系中，本文对试题模型ontolog、染色体的结构、以及适应度函数进行了合理设计，同时给出所有参与agent任务的详细设计及其状态图、顺序图和活动图。 最后，为了验证本文所提出的体系、模型和方法的可行性和有效性，我们运用一个与fipa兼容的agent中间件─jade对部分模型进行了仿真和实现。首先，为了演示如何具体实现一个大规模的多agent系统，我们选择对gamastp原型进行了具体实施，这里面涉及到如何运用设计ontology, 如何实现agent之间的通讯，如何设计agent行为以及如何在分布式环境中分布配置各种agent；其次，为了展示如何实现fipa的交互协议，我们选择用户模型agent作为试验对象；最后，我们部分实现了本文提出的对等帮助系统模型。所有的仿真结果证明了本文所提出模型是可行的和有效的。 Nowadays there is an increasing demand for the e-Education market since the e-Education paradigm has the potential to revolutionize the way of learning by making it individual rather than institution-based, eliminating clock-hour measures in favor of performance and outcome measures, and emphasizing customized learning solutions rather than one-size-fits-all instruction regardless of geographical, temporal, physical, social, and economical constraints. Typically, an e-Education system is a high dynamical, open, unpredicted from system engineering view. in such setting, in addition to the common features of distributed systems such as concurrency, distributed, hypermedia, etc., the e-Education systems have the new features of autonomy, evolutionary life-cycle, collaboration, etc. Consequently, software engineering of such systems is confronted with a number of challenges, such as to deal with service-oriented computing, dynamic integration of autonomous components, distributed and mobile computing, etc. Although the benefits and potentials of the new generation of e-Education are obvious and exiting, unfortunately, so far the great potential of e-Education has been far from being taken full of advantage. Currently most of today’s e-Education systems are dominated by the objectivist school and the use of technology as a substitute for a teacher delivering instruction. Current approaches to the online learning environment usually transfer traditional classroom instruction to an online setting, recasting reading materials as web-based materials. Apparently, the current e-Education does not seem to fulfill its promise to become the most important learning paradigm, especially in the context of the increased role of continuous and life-long learning. From the learner perspective, they often complain about the lack of flexible performance tools in support of personalized and tailored learning, value-added reflection, mutual simulative knowledge sharing, on-demand expertise finding, just-in-time peer help as well as efficient and timely tutor guidance. From the tutor perspective, the main drawback of current e-Education systems is that they tend to require more effort in terms of authoring learning materials and preparing tests or examinations than their classical counterparts. The necessity of mastering technology-intensive teaching tools and the lack of the tutor’s computer literacy often make tutors reluctant to participate in online teaching activity. Consequently, it is obvious that on one hand, we need to provide learner with more intelligent learning environment that supports various customized learning services as needed, on the other hand, we need innovative mechanism to alleviate tutor workload in terms of facilitating the development of learning contents and test//exam by hiding as much technique details as possible. To address these issues mentioned above, in this dissertation, we launched a joint initiative named MAGE between HUST and ENIM under the support of DUO-France The eventual goal is develop an intelligent, flexible, personalized and open e-Education environment in order to provide an efficient mechanism to personalize the learner’s learning process and the teacher’s pedagogic process, diversify the learning paradigms and facilitate the development of teaching and learning materials. To achieve such goal, we explored, and adopted a series of innovative methodologies, theories, algorithms, and technologies derived from multiple disciplines such as Multi-Agent System /(MAS/), Learning Object /(LO/), Cognitive Theory /(CT/), Genetic Algorithm /(GA/), eXtensible Markup Language /(XML/), J2EE and so on. In particular, we, in this dissertation, concentrate on the approach of MAS as a container and supporting environment to integrating and encapsulating the above mentioned technologies and methodologies, as well as to modeling and implementing several typical e-Education applications at different levels and different contexts in terms of content authoring, individual and collective learning, expertise peer help finding, and test generation, delivery, assessment in distributed learning environment after deliberately taking into consideration the obvious advantage of MAS in terms of both its property such as autonomy, proactiveness, social ability and reactivity, and its distinctive features with regard to modularity, abstraction, parallel computation, robustness, scalability, legacy systems encapsulation, reliability, extensibility, robustness, maintainability, flexibility and reusability. To efficiently built an open, adaptive and personalized MAS based e-Education system, this thesis proposed an LTSA-compliant and MAS-based e-Education architecture—MAGE. This architecture consists of numerous agents, which perform different specific tasks on behalf of different learners, resources, applications or even computing by cooperation, negotiation, communication among them. In this way, we achieved a rather complex, dynamical, open, self-organizing and adaptive multi-user, multi-agent based e-Education system. In particular, our focus is concentrated on the following aspects: In the domain of the course authoring, the key issue is how to develop instructional materials of high quality that could be reused and applied to different contexts. Unfortunately, these instructional contents are, traditionally, expensive and time consuming to produce. To this problem, this thesis put forward an architecture of multi-agent enabled course authoring model based on e-Education object /(MEEOCAS/), involving the proposition of the concept model and the definition, structure and package model of the EEO. Under support of this subsystem, the course designers may conveniently develop their courses through assembling the ready-made learning objects /(i.e., EEOs/) instead of creating them from the scratch. To enhance the flexibility, this framework also supports several services such as subscription, searching, and registration and publishing of learning objects. As far as adaptive and personalized learning is concerned, this thesis proposed a MAS based integrated framework in support of adaptive and active learning in both individual and collective learning spaces. The distinct advantage of the proposed framework consists in the efficient integration of the two adaptive mechanisms by virtue of the cooperation, negotiation and communication among multi agents. In the adaptive individual space, the key issue is how to dynamically generate personalized learning path consisting of domain concepts and present associated learning objects catering for a learner’s knowledge state and learning preference. As to this, this thesis put forward an efficient searching algorithm for the presentation generation based on the proposed domain ontology model. In the collective learning space, our focus is on the issue how to find appropriate help resources /(e.g. peer learners, learning materials, or other applications/) and how to dynamically build a tailored learning group on behalf of learners in a distributed network according to their need. In this regard, this thesis proposed two corresponding architectures: One is the peer help system, another is architecture of the learning group forming system, in which, individual learners can establish a .collaboration profile. indicating the characteristics of the group they would like to participate. The proposed collective learning architecture is based on several agents which perform functions such as seeking for potential collaboration partners, expressing which collaboration services are to be used, and monitoring collaborative learning activities. With regard to the e-assessment, the traditional computer based evaluation mechanisms rely predominately on the client-server model. Such mechanisms usually do not scale well and do not fully support features like automatic test generation, evaluation of subjective questions, delivery of dynamic content, off-line examinations, flexible communication between online evaluation components, and proactive event notification etc. to address such issues, this thesis put forward an innovative holistic solution to modeling large-scale on-line assessment system by applying the new generation of mobile agent based distributed computing paradigm. In particular, the most significant innovative point consists in that we proposed and designed an innovative model of automatic test generation by seamlessly integrating genetic algorithm, mobile agent, and MAS. Since mobile agents are autonomous and dynamic entities that have the ability to migrate between various nodes in the network, they offer many advantages over traditional design methodologies like reduction in network load, overcoming network latency and disconnected operations etc. Eventually, in order to verify and validate the feasibility and efficiency of the models proposed in this thesis, we implemented and simulated part of the models with the JADE framework. especially, we implemented three typical applications proposed in this thesis. First, we implemented the simplified prototype of GAMASTP for the purpose of synthetically revealing how to concretely implement a complex multi-agent system, which is concerned with several key issue: how to implement test ontology and apply to the communication among agents; how to design and implement agent behavior model according to the previous models; how to deploy agents over different network nodes. the second application is implemented for the purpose of how the learner model agent updates the learner model upon receiving the refresh data as well as how to answer any questions from external agents, this example also showed the application of interactive protocols such as FIPA request and FIPA query. The third example is used to implement part of the peer help system aiming at demonstrating the process how to find appropriate competent peer learners. The simulation results show the feasibility and efficiency of the models proposed in this dissertation.

关 键 词： 远程教育 学习对象 自适应学习 在线考试 遗传算法 系统集成 构建主义

分 类 号： [G434]

领　　域： [文化科学] [文化科学]