融合多层注意力机制与知识图谱的课程推荐OA

Personalized course recommendations have become an important technology to solve the problem of information over-load and improve the learner's experience.However,existing methods often fail to effectively exploit the multi-dimensional in-formation embedded in learning behaviors and suffer from issues such as data sparsity and poor performance under cold-start con-ditions.To address these limitations,this paper proposes a novel course recommendation model(MAKR),which combines a multi-layer attention mechanism with knowledge graph embedding.Specifically,the proposed model leverages a multi-layer at-tention module to extract rich,multi-dimensional features from four heterogeneous data sources,including learning records,course descriptions,user reviews,and temporal learning nodes.This design enables the model to capture the dynamic changes of user interests and uncover the deep preferences behind users'composite behaviors.Furthermore,a domain-specific knowledge graph is constructed,containing 21757 entity nodes and 116557 triples.Knowledge graph embedding is employed to model the structured relationships among courses,and the resulting representations are incorporated as auxiliary information alongside user interest features.This strategy effectively mitigates data sparsity and cold-start issues while enhancing the interpretability of rec-ommendation results.Extensive experiments conducted on the MOOCCubeX dataset demonstrate that,compared with baseline methods,the proposed model improves Recall@5 by 7.63%and accuracy by 1.41%.Moreover,in sparse scenarios,the area un-der the ROC curve decreases by only 5.47%,indicating strong robustness.These results validate the effectiveness and superiority of the proposed approach.