利用维基百科点击流的概念依赖关系识别方法

Identifying Prerequisite Relations Between Concepts Using the Wikipedia Clickstream

学习资源中的知识概念间的依赖关系决定了每个资源在整个课程中学习顺序.当前主流的概念依赖关系挖掘方法是利用概念自身的特征进行依赖关系的识别.而用户在学习资源中的点击操作日志同样能够预示概念间依赖关系的存在.本文以维基百科为例,提出一种基于点击流的概念依赖关系识别方法,利用维基百科中用户与词条页面的交互信息,设计相关概念集合特征来预测概念间的依赖关系,在实验过程中,为了使分类器能够充分学习不同类别数据,使用过采样方法使训练集类别平衡,观察各度量指标在不同机器学习分类器上的表现,选取最佳的分类器进行后续实验.实验结果表明,本文的方法与当前研究方法相比,既保证了维基百科原始数据具有较高的概念对覆盖率,也能够有效的预测概念间的依赖关系.

The prerequisite relations between knowledge concepts in learning resources determine the learning order of each resource in the entire course.The current mainstream identifying prerequisite relations between concepts methods are to use the characteristics of the concept itself to identify the prerequisite relations.The user′s click operation log in the learning resources can also predict the existence of prerequisite relations between concepts.This paper proposes a method for identifying prerequisite relations based on clickstreams,using the interaction information between the user and the entry page in Wikipedia,designing set of related concepts features to predict prerequisite relations among concepts,based on Wikipedia.During the experiment,in order to enable the classifier to fully learn different types of data,this paper uses the oversampling method to balance the training set categories,observing the performance of each metric on different machine learning classifiers and selects the best classification for subsequent experiments.Experimental results show that compared with current research methods,the method in this paper not only ensures that the original Wikipedia has a high concept-to-coverage rate,but also can effectively predict the prerequisite relations between concepts.