基于矩阵分解的多模态军事职业教育在线学习成果预测

摘要：

军事职业教育的在线学习成果预测因其教育意义而受到广泛关注。在线学习成果预测方法中，基于推荐的方法凭借其突出的个性化优势占有重要地位。虽然现有研究通过利用辅助信息来解决推荐系统中冷启动和合并侧面信息的问题，但基于多模态推荐在线学习成果预测方法的潜力尚未研究。首次使用多模态辅助学习者在线学习成果预测，使用大语言模型作为信息混合器，为矩阵分解产生额外的引导信号。具体来说,所提的语言引导矩阵分解模型，使用语言化的多模态信息，为教育交互产生丰富的语义嵌人，并将它们作为矩阵分解的辅助信号。提供了两种文本嵌入方法，嵌入初始化方法计算数据的特定先验以初始化分解矩阵，嵌入蒸馏方法用于对齐矩阵分解的潜在特征和嵌人特征，使语言模型更充分地指导矩阵分解。在一个军事职业教育平台收集的数据集上评估了所提出的模型，包括 2.5万个学习者的 35万个互动和5万个不同的成果表现。大量实验表明，当前模型在各种军事职业教育在线学习成果预测场景，包括冷启动中优于已有方法。

Online learning outcome prediction in military vocational education has received wide attention due to its educational significance, in online learning outcome prediction methods, recommendation-based methods are more important. However, the recommendation svstem has problems with cold-start and incrporating side information, Although the existing work addresses the above problems by utilizing auxiliary information, the potential of multi-modal based recommended online learning outcome prediction methods has not been investigated, For the first time, this paper uses multi-modal assisted student onlime learning outcome prediction, using a large language model as an information mixer to generate additional guidance signals for matrix factorization, Speciically, the language guided matrix factorization model in this paper, using linguistic multi-modal imformation, produces rich semantic embeddings for educational interactions and uses them as auxiliary signals for matrix factorization. This paper provides two text embedding methods, embedding initialization method calculates specific priors of the data to initialize the decomposition matrix, and embedding distillation method is used to align the underlying features and embedding features of the matrix decomposition, making the language model more fully guide the matrix decomposition, This paper evaluates the proposed model on an online education platform, including 3.5 X10₅ interactions for 2.5X10₄ users and 5X10₄ diferent evaluations of outcomes, A lot of experiments show that the current model outperforms the existing methods in various military vocational education online learning outcome prediction scenarios, including cold-start.

作者:

邵东春,李康,杜英,赵涵

Shao Dongchun，Li Kang，Du Ying，Zhao Han

机构地区：

陆军装甲兵学院蚌埠校区

引用本文：

邵东春，李康，杜英等。基于矩阵分解的多模态军事职业教育在线学习成果预测[J].河南师范大学学报(自然科学版),2025,53(6):58-65.(Shao Dongchun,Li Kang, Du Ying,et al. Multi-modal military vocational education online learning outcome prediction based on matrix factorization[J].Journal of Henan Normal University(Natu-ral Science Edition),2025,53(6):58-65.DOI:10.16366/j.cnki.1000-2367.2024.08.12.0002.)

基金：

军队科研资助重点项目；2024年度全国教育科学国防军事教育学科规划军队重点课题

关键词：

军事职业教育;在线学习成果预测;多模态;冷启动;矩阵分解

military vocational education; online learning outcomes predication; multi-modal; cold-start; matrix factori-zation

分类号：

TP315.69

基于矩阵分解的多模态军事职业教育在线学习成果预测.pdf