Journal of Frontiers of Computer Science and Technology ›› 2024, Vol. 18 ›› Issue (8): 2049-2064.DOI: 10.3778/j.issn.1673-9418.2305069

• Theory·Algorithm • Previous Articles     Next Articles

MOOCs Knowledge Concept Recommendation Model Based on PathSim

ZHU Yi, JU Chengcheng, HAO Guosheng   

  1. 1. School of Computer Science and Technology, Jiangsu Normal University, Xuzhou, Jiangsu 221116, China
    2. Educational Informatization Engineering Technology Research Center, Xuzhou, Jiangsu 221116, China
    3. Key Laboratory of Safety-Critical Software (Nanjing University of Aeronautics and Astronautics), Ministry of Industry and Information Technology, Nanjing 211106, China
  • Online:2024-08-01 Published:2024-07-29

基于PathSim的MOOCs知识概念推荐模型

祝义,居程程,郝国生   

  1. 1. 江苏师范大学 计算机科学与技术学院,江苏 徐州 221116
    2. 江苏省教育信息化工程技术研究中心,江苏 徐州 221116
    3. 高安全系统的软件开发与验证技术工业和信息化部重点实验室(南京航空航天大学),南京 211106

Abstract: Massive open online courses play a crucial role in advancing modern education by providing extensive open online learning platforms. However, there are still challenging aspects to consider when it comes to reducing user learning blind spots and improving the overall user experience. Firstly, interaction data are sparse. Secondly, scaling up to large-scale recommendation tasks is difficult. Thirdly, user needs are not solely determined by individual preferences, but are also influenced by different teachers and course materials. Fourthly, developing a unified model that can effectively represent different types of entities and relationships within course learning events is a challenging task. This paper introduces a relevance metric that computes the weights of edges by leveraging the structural information of entire graph. This paper presents the PathSimSage model (path-based similarity sampler and aggregate) for recommending knowledge concepts, utilizing the PathSim algorithm (path-based similarity) for neighborhood sampling. The relevance scores between entities are precomputed offline, which decouples the neural network from the propagation process. This decoupling maintains the independence of the network??s layered architecture from the propagation mechanism, thereby considerably reducing the training time of model. Through extensive experimentation on the publicly accessible MoocCube dataset, PathSimSage has shown to minimize the impact of irrelevant or noisy information, resolve the significant node bias induced by random walk sampling, and somewhat alleviate the issue of oversmoothing.

Key words: massive open online courses, graph neural networks, personalized course recommendations, graph convolution, metapath-based subgraphs, similarity measure

摘要: 大规模开放在线课程提供大规模开放式在线学习平台,为推进现代教育发挥关键作用。然而,减少用户学习盲区和改善用户体验方面的研究仍具有挑战性:交互数据稀疏;难以扩展到大型推荐任务上;用户需求不单由用户喜好决定,还受到不同教师、课程影响;以统一的方式对课程学习事件中不同类型实体及关系进行建模并不妥靠。基于此,引入相关性度量,依据全图结构信息计算各边权重,提出采用相关性度量算法PathSim进行邻域采样的知识概念推荐模型PathSimSage。各实体间相关性得分可在本地离线计算,将神经网络与传播过程分离,保证神经网络的堆叠层数和传播过程的独立性,大幅减少模型所需训练时间。在公开的MoocCube数据集上进行了综合实验,PathSimSage降低了不相关的信息甚至噪声的影响,解决随机游走采样所引发的高度节点偏差问题,并在一定程度上缓解了过平滑效应。

关键词: 大规模开放在线课程, 图神经网络, 个性化课程推荐, 图卷积, 基于元路径的子图, 相似性度量