面向高校课程的知识图谱联合嵌入模型研究

Research on the Joint Embedding Model of Knowledge Graphs Oriented to University Curriculums

知识图谱嵌入是构建高效精准知识图谱的基础,但目前大多数知识图谱嵌入方法只注重结构信息,而忽视或未能充分利用实体的背景信息,使得嵌入信息不够准确。为充分利用课程实体中丰富的信息以更好地将知识图谱技术应用于教育场景,文章设计了一种面向高校课程的知识图谱联合嵌入模型,包含结构信息嵌入、目录信息嵌入、语义约束计算、联合嵌入四个模块。之后,文章进行了链接预测实验和实体分类实验,来分别验证该模型的有效性和准确性,实验结果显示:该模型的性能良好;与现有模型相比,该模型嵌入的质量和分类的效果均得到较大提升。最后,文章展望该模型可为面向课程的智能问答、路径推荐、信息检索和数据的可视化交互等下游教育应用提供有力支撑,其研究将加快推动知识与数据驱动的智能化教育落地。

The embedment of knowledge graphs is the foundation for the construction of efficient and accurate knowledge graphs.However,most knowledge graph embedding methods only focus on structural information,but neglect or fail to make full use of entity background information,resulting in the inaccuracy of embedding information.In order to make full use of the abundant information in the curriculum entity and better apply the knowledge graph technology to the educational scene,the joint embedding model of knowledge graphs oriented to university curriculums was designed in this paper,which included four modules of structure information embedding,directory information embedding,semantic constraint calculation,and joint embedding.Then,the link prediction experiment and the entity classification experiment were conducted to verify the validity and accuracy of the model.Experimental results showed that the model had good performance,and the embedding quality and the classification effect of this model were greatly improved,as compared with the existing models.Finally,the strong support of the model for downstream education applications such as intelligent question answering,path recommendation,information retrieval and visual interaction of data oriented to curriculums was expected,and the research of this paper could accelerate the implementation of knowledge and data-driven intelligent education.