基于多任务自编码器的MOOC课程推荐模型  

MOOC course recommendation model based on multi-task autoencoder

作　　者：董永峰 王巍然[1] 董瑶 史进 王雅琮[1] DONG Yong-feng;WANG Wei-ran;DONG Yao;SHI Jin;WANG Ya-cong(School of Artificial Intelligence,Hebei University of Technology,Tianjin 300401,China;Engineering Research Center of Integration and Application of Digital Learning Technology,Ministry of Education,The Open University of China,Beijing 100039,China;Hebei Province Key Laboratory of Big Data Computing,Hebei University of Technology,Tianjin 300401,China)

机构地区：[1]河北工业大学人工智能与数据科学学院,天津300401 [2]国家开放大学数字化学习技术集成与应用教育部工程研究中心,北京100039 [3]河北工业大学河北省大数据计算重点实验室,天津300401

出　　处：《计算机工程与设计》2023年第10期3117-3123,共7页Computer Engineering and Design

基　　金：河北省高等学校科学技术研究基金项目(QN2021213);数字化学习技术集成与应用教育部工程研究中心创新基金项目(1221006)。

摘　　要：为解决在线学习当中,学习者行为的数量远少于在线课程的样本总数所产生的数据稀疏问题,提出一种基于多任务自编码器的课程推荐模型(multi-task autoencoder course recommendation model,MAEM)。通过分析学习者的学习行为,将总体任务分为两个子任务:任务一是学习者浏览课程章节列表行为,任务二是完成课程50%的学习行为,通过共享网络底部的隐藏层提高泛化能力。模型总体划分为共享嵌入、自编码器与分解预测、任务组合3个模块,3个模块协同工作,旨在突破训练数据稀疏问题。将MAEM与7种常用的推荐算法比较,实验结果表明,MAEM算法优于7种热门的推荐算法,验证了其在课程推荐中的有效性。There is a big difference of the amount of data between learners’behaviors and the courses on online education.To solve the problem of data sparsity,a multi-task autoencoder course recommendation model(MAEM)was proposed.By making full use of the behavior characteristics of students’online learning,the overall task was divided into two sub-tasks.The first task was the behavior of browsing the list of course chapters,and the second task was the behavior of finishing 50%of the course.These two sub-tasks were sharing with the hidden layer at the bottom of the network to improve the generalization ability.The model was generally divided into three modules including shared embedding,decomposition prediction and autoencoder,and task combination.These three modules worked together to break through the problem of sparse training data.Through the experiments,the proposed model outperforms seven state-of-the-art ones and provides more accurate course recommendations for students.

关 键 词：多任务学习 自编码器 课程推荐 数据稀疏性 行为分解 特征提取 在线学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]