一种基于HGNN与多层感知机的网络职业教育推荐方法

A recommendation method of network vocational education based on HGNN and multi-layer perceptron

摘要

随着网络职业教育的快速普及,平台亟需为学习者提供精准的个性化推荐服务。然而,传统推荐系统在处理海量、高维度且关系复杂的教育数据时存在明显不足。针对此问题,提出了一种基于超图神经网络(hypergraph neural network,HGNN)与多层感知机(multilayer perceptron,MLP)相融合的网络职业教育推荐方法。该方法首先构建综合多种实体节点的异构图模型,利用HGNN的高阶超边结构捕捉教育数据中的复杂关系;随后,引入多层感知机对提取的特征进行非线性强化学习,以精准预测用户与学习资源的链接概率。基于真实教学平台数据的实验结果表明,在推荐列表长度设为10项时,本模型命中率达到最优;随着训练的推进,系统准确率由88%显著提升至96%,整体性能超越了传统矩阵分解等方法。参数敏感性分析进一步验证了模型对关键参数响应的鲁棒性。本研究有效提升了推荐系统的准确性与稳定性,为网络职业教育平台的个性化学习服务提供了新的优化方向与技术支撑。

Abstract

With the rapid popularization of online vocational education, platforms urgently need to provide learners with accurate personalized recommendation services. However, traditional recommendation systems have obvious shortcomings in processing massive, high-dimensional, and complex educational data. To address this issue, this study proposes a network vocational education recommendation method based on the fusion of Hypergraph Neural Network (HGNN) and Multilayer Perceptron (MLP). This method first constructs a heterogeneous graph model integrating multiple types of entity nodes, and uses the high-order hyperedge structure of HGNN to capture complex relationships in educational data. Subsequently, an MLP is introduced to conduct nonlinear reinforcement learning on the extracted features to accurately predict the link probability between users and learning resources. Experimental results based on real teaching platform data show that when the recommendation list length is set to 10 items, the model′s hit rate reaches the optimal level. As training progresses, the system′s accuracy significantly improves from 88% to 96%, surpassing traditional methods like matrix factorization. Parameter sensitivity analysis further verifies the robustness of the model′s response to key parameters. This study effectively enhances the accuracy and stability of recommendation systems, providing a new optimization direction and technical support for personalized learning services in online vocational education.

关键词

职业教育推荐 / 超图神经网络 / 多层感知机 / 个性化学习 / 异构图模型 / 推荐系统

Key words

vocational education recommendation / hypergraph neural network / multilayer perceptron / personalized learning / heterogeneous graph model / recommendation system

[1] 杨菁, 张欢, 张烨宇, 等. 基于双池化的多层感知机光功率预测方法[J]. 城市轨道交通研究, 2024, 27(4):247-251, 261.
[2] 王谢中, 陈旭, 景永俊, 等. 基于异构图神经网络的半监督网站主题分类[J]. 计算机工程与科学, 2024, 46(4):635-646.
[3] Dahalan F, Alias N, Shaharom M S N. Gamification and game based learning for vocational education and training: a systematic literature review[J]. Education and Information Technologies, 2024, 29(2): 1279-1317.
[4] 徐胤, 王殊, 陈鹏, 等. 基于人工智能AI大模型的电力系统运行控制方法[J/OL]. 自动化技术与应用, 1-5[2026-05-26]. https://link.cnki.net/urlid/23.1474.TP.20241227.1558.100.
[5] 李荣远, 龙法宁. 多核超图神经网络单细胞分类算法[J]. 计算机应用与软件, 2023, 40(9):303-307, 340.
[6] 秦志龙, 邓琨, 刘星妍. 基于元路径卷积的异构图神经网络算法[J]. 电信科学, 2024, 40(3):89-103.
[7] 马亚杰, 刘洋, 姜斌, 等. 基于异构图注意力的工业设备故障诊断知识图谱补全方法[J]. 中国科学:信息科学, 2024, 54(2):354-372.
[8] 林晶晶, 冶忠林, 赵海兴, 等. 超图神经网络综述[J]. 计算机研究与发展, 2024, 61(2):362-384.
[9] 李彤, 阎宇航, 安静, 等. 基于触觉阵列感知的机器人稳定抓取判别方法[J]. 华中科技大学学报(自然科学版), 2024, 52(5):136-143.
[10] 李挺, 金福生, 李荣华, 等. Light-HGNN:用于圈层内容推荐的轻量同质超图神经网络[J]. 计算机研究与发展, 2024, 61(4):877-888.
[11] 周帅, 刘新, 李真, 等. 融合多源海洋大地测量数据的南海海底地形多层感知机反演[J]. 地球物理学报, 2024, 67(4):1368-1382.
[12] 彭勃, 李耀东, 龚贤夫, 等. 一种基于异构图神经网络和文本语义增强的实体关系抽取方法[J]. 计算机科学, 2024, 51(增刊1):268-272.
[13] 唐佳, 贾燕, 史绍恩, 等. 智能教育下考虑不同教育场景的线上内容推荐算法研究[J]. 自动化技术与应用, 2025, 44(7):99-103.
[14] 李红春. 基于物联网技术的电子学习交流系统[J]. 自动化技术与应用, 2024, 43(2):139-141.
[15] 任俊, 黄瑞章. SSHGCN:基于音形异构图卷积的中文纠错方法[J]. 山西大学学报(自然科学版), 2024, 47(3):518-527.

基金资助

陕西省职业技术教育研究项目(ZJS202131)