摘要：

在现有的教学资源推荐系统中算法和模型不断进步, 但推荐策略仍然停留在学生偏好分析或内容相关性计算两方面, 忽略了教学资源推荐中授课教师对教学目标的掌控以及企业导师的实践价值引导作用。为此, 提出一种基于图神经网络和多主体评价的教学资源推荐方法, 综合考虑学生、教师和企业导师3个主体的评价, 并解决多主体评价导致的小样本问题。首先, 构建包含教学资源和知识点的异质图, 通过异质图嵌入分别获取教学资源和知识点的特征向量表征, 计算教学资源与知识点的相关性; 然后, 设计多主体评价机制, 分别制定针对学生、教师和企业导师3个不同主体的细粒度评价指标, 利用基于图神经网络的小样本学习(GNN-FSL)模型获取不同主体对教学资源的评分向量估计; 最后, 利用注意力机制来综合不同主体的评分向量估计以及教学资源与知识点的相关性对推荐结果的影响。实验结果表明, 该方法在小样本训练中具有较好的评分向量估计准确度优势, 并且融合多个主体评价和增加细粒度多主体评价指标的两大措施均对提升推荐准确度和学习成绩具有显著作用。

关键词: 教学资源推荐, 图神经网络, 评价, 多主体, 小样本

Abstract:

In existing education resource recommendation systems, an increasing number of novel algorithms and models have been developed, consistently considering user preference analysis and content relevance calculation. However, these systems often neglect the influence of education objective plans from teachers and practical guidance from corporate mentors. Therefore, this study proposes a education resource recommendation method based on Graph Neural Network (GNN) and multi-subject rating, which considers ratings from multiple subjects, including students, teachers, and corporate mentors, further solving the few-shot problem caused by the multi-subject rating. In this method, education resources and knowledge points are first put into a heterogeneous graph and then their feature vectors are represented via heterogeneous graph embedding to calculate the relevance between them. Subsequently, a multi-subject rating mechanism is designed, in which fine-grained rating indicators are defined for different subjects and the corresponding rating values from different subjects are estimated via the GNN-based Few-Shot Learning (GNN-FSL) model. Finally, an attention mechanism is utilized to integrate the influence of rating values from different subjects and the relevance of education resources and knowledge points on the recommendation results. The experimental results indicate that the proposed method not only improves rating accuracy in few-shot training but also facilitates multi-subject rating and fine-grained rating indicators to improve both recommendation accuracy and student grades.

Key words: education resource recommendation, Graph Neural Network(GNN), rating, multi-subject, few-shot