基于CVGAE的无监督跨领域学习先决条件链挖掘  

Unsupervised Cross-Domain Learning Prerequisite Chain Mining Based on CVGAE

作　　者：徐国兰[1] Xu Guolan(Library of Shandong University of Technology,Zibo 255049)

机构地区：[1]山东理工大学图书馆,淄博255049

出　　处：《现代计算机》2022年第23期65-69,共5页Modern Computer

摘　　要：在学习过程中,概念之间存在一种先决条件关系,概念先决条件链挖掘可以帮助人们提高获取新知识的效率。依靠大规模标注数据,人们可以通过监督学习快速确定概念的学习顺序。但是在没有标注数据的领域,先决条件关系很难被发现,只能借助其他领域有标注的数据进行跨领域迁移。采用无监督跨领域变分图自编码器(CVGAE)可以解决跨领域先决条件链发现问题。CVGAE模型由变分图自编码器和领域判别器组成,变分图自编码器用来预测概念之间是否具有先决条件关系,领域判别器用来判断概念所属领域。模型只需要简单的同构图作为输入,不需要构建复杂的异构图。通过在自建的数据集上评估,相较于目前最先进的基于资源概念图的基准模型,该模型在仅使用20%的图规模和35%的计算时间的情况下却能够取得相当的效果。In our learning process, there is a prerequisite relationship between concepts. Conceptual prerequisite chain mining can help people improve the efficiency of acquiring new knowledge. People can quickly determine the order in which concepts are learned through supervised learning relying on large-scale labeled data. However, in the field without labeled data, it is difficult to find the prerequisite relationship, and cross-domain migration can only be carried out with the help of labeled data in other fields. Employing an unsupervised cross-domain variational graph autoencoder(CVGAE)can solve the cross-domain prerequisite chain discovery problem. The CVGAE model consists of a variational graph autoencoder and a domain discriminator: the variational graph autoencoder is trained to predict preconditional relationships between concepts, and the domain discriminator is used to determine the domain to which a concept belongs. This method only requires simple isomorphic graphs as input, and does not need to construct complex heterogeneous graphs. Compared with the current state-of-the-art benchmark models based on resource concept maps, it is evaluated on a self-built dataset. The model achieves comparable results using only 20% of the graph scale and35% of the computation time.

关 键 词：先决条件链 跨领域 变分图自编码器 深度学习 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]