基于自动化知识网络的知识导航与阻塞效应研究

发布时间：2018-03-09 05:10

本文选题：自动化知识网络　切入点：复杂网络　出处：《太原理工大学》2016年硕士论文　论文类型：学位论文

【摘要】：知识网络概念随着知识经济的兴起,逐渐渗透到社会学、经济学、情报学、管理学、教育学等诸多领域。学科知识网络就是知识网络在教育学领域内的一个应用,它已经发展成为一种有效的工具,用以直观地描述具备复杂非线性结构的学科知识体系。针对现阶段大学生在学习过程中容易忽略知识联系的问题,学科知识网络可以明确知识联系并为知识导航技术提供基础平台。相比于网络学习环境下的近似导航路径,清晰的知识关联结构与合理的学习顺序更有益于提升学习者对学科专业知识的学习效果。因此本文借助复杂网络的分层递阶结构分解方法和有向图的相关理论基于学科知识网络推荐学习路径,实现知识导航并指导学习者构建个人知识网络。为了衡量知识节点的不同属性及知识关联对学习过程的重要性,本文以知识节点的通用连接结构为研究对象,通过真实的成绩数据利用函数拟合与云模型两种方法逼近先学知识节点与后继知识节点间的非线性激励关系。将可靠性较高的激励关系用来评估先学知识掌握程度对后继知识掌握程度的影响。最后将这个激励关系应用在学科知识网络上进行知识阻塞实验,探究个别知识掌握程度不理想对整体学习效果的影响。为自动化学科知识体系构建相应学科知识网络,并挖掘该知识网络在个人学习及教学中的实际应用意义。本文具体工作如下:(1)确定自动化知识网络构建过程中的相关原则,用以约束知识节点的选取及知识节点关联的构建工作。对知识网络进行建模和描述,形成以课程、知识点为节点,以其先学关系为连边的两层自动化知识网路。(2)利用Pajek软件对知识网络进行可视化。借助复杂网络理论统计网络基本特性,统计结果显示节点的度值、聚集系数、介数均可以用来挖掘重点知识。(3)基于知识点网络的拓扑信息,采用复杂网络的分层递阶分解理论计算可达性矩阵,从而减小知识搜索空间,并通过冗余路径归并和并行拓扑排序算法推荐学习路径以实现知识导航。(4)对知识学习过程中涉及的相关问题进行了基本假设与描述。利用成绩数据,以数据拟合和云模型两种方法逼近先学知识与后学知识间的非线性激励关系。实验结果表明,针对该问题云模型对离群点的容忍能力较强且具有较高的逼近能力。(5)将云模型逼近的非线性激励关系引入知识点网络,加入不同的变量因素进行知识阻塞实验。实验结果表明当掌握情况不理想的知识节点的出度值较大、位于网络中的层数较低或者所处网络先学连边较稀疏时都会明显削弱学习者的整体学习效果。将上述实验结果与自动化方向的个人知识学习过程和教学实际相结合,将会对提升学习者的学习效率和指导教学产生积极作用。
[Abstract]:With the rise of knowledge economy, the concept of knowledge network has gradually penetrated into many fields, such as sociology, economics, information science, management science, pedagogy and so on. It has developed into an effective tool to describe the subject knowledge system with complex nonlinear structure intuitively. The subject knowledge network can define the knowledge connection and provide the basic platform for the knowledge navigation technology. Compared with the approximate navigation path in the network learning environment, A clear knowledge association structure and a reasonable learning sequence are more beneficial to improve the learning effect of the subject and specialty knowledge. Therefore, this paper makes use of the hierarchical structure decomposition method of complex network and the related theoretical basis of directed graph. Recommend the learning path to the subject knowledge network, In order to measure the importance of different attributes of knowledge nodes and the relevance of knowledge to the learning process, this paper takes the common connection structure of knowledge nodes as the research object. The nonlinear excitation relationship between the first learning knowledge node and the subsequent knowledge node is approximated by the real achievement data by two methods of function fitting and cloud model. The high reliability incentive relation is used to evaluate the first learning knowledge mastery process. The influence of degree on the degree of subsequent knowledge mastery. Finally, the incentive relationship is applied to the knowledge blocking experiment on the subject knowledge network. To explore the influence of individual knowledge acquisition degree on the overall learning effect. To build the corresponding knowledge network for the automation subject knowledge system, And excavate the practical application significance of the knowledge network in personal learning and teaching. The specific work of this paper is as follows: 1) to determine the relevant principles in the process of constructing the automated knowledge network. It is used to constrain the selection of knowledge nodes and the construction of knowledge nodes association. The knowledge network is modeled and described to form courses and knowledge nodes. The two-layer automated knowledge network with its first learning relation as the connecting edge) visualizes the knowledge network by using Pajek software. With the aid of the complex network theory, the basic characteristics of the network are analyzed, and the statistical results show the degree value and aggregation coefficient of the nodes. All of them can be used to mine the key knowledge. 3) based on the topological information of knowledge point network, the hierarchical decomposition theory of complex network is used to calculate the reachability matrix, so as to reduce the search space of knowledge. The redundant path merging and the parallel topological sorting algorithm are used to recommend the learning path to realize the knowledge navigation. The related problems in the process of knowledge learning are hypothesized and described. Two methods, data fitting and cloud model, are used to approximate the nonlinear excitation relationship between the first knowledge and the later knowledge. The experimental results show that, To solve the problem, the cloud model has a strong tolerance to outliers and has a higher approximation ability.) the nonlinear excitation relation of cloud model approximation is introduced into the knowledge point network. The experiment results show that the outlier of the knowledge node is higher than that of the unsatisfied knowledge node when different variable factors are added to the experiment. When the number of layers in the network is lower or the learning edge of the network is sparse, the overall learning effect of the learner will be significantly weakened. The above experimental results are combined with the automated direction of personal knowledge learning process and teaching practice. It will play a positive role in improving learners' learning efficiency and guiding teaching.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：G434;O157.5

【相似文献】