知识状态预测与聚类驱动的习题推荐方法

Knowledge State Prediction and Cluster-Driven Exercise Recommendation

摘要: 个性化习题推荐是提升学习效率、实现精准教学的关键。传统习题推荐多依赖协同过滤范式，忽略了学生知识状态与习题难度的动态关联，推荐结果针对性不足。基于知识追踪的方法虽能结合知识掌握程度匹配习题，但仍存在局限：一是对学生、习题、知识点的关联建模不足，忽视细粒度答题信息，降低了知识状态预测与推荐精度；二是未充分利用学生群体的知识共性，难以提升推荐多样性与新颖度。为此，本文提出融合图知识追踪与聚类驱动的习题推荐方法KSPC-ER。该方法采用双图知识追踪精准刻画学生知识状态，借助K-means++挖掘群体学习共性，并设计多技能关联难度计算与动态推荐阈值，实现习题精准适配。实验结果表明，该方法在推荐准确率、新颖性和多样性上均优于现有基线模型，有效性与优越性得到验证。

Abstract: The Personalized exercise recommendation is the key to improving learning efficiency and realizing precision teaching. Traditional exercise recommendation methods mostly rely on the collaborative filtering paradigm, which ignores the dynamic correlation between students’ knowledge states and exercise difficulty, leading to insufficient pertinence of recommendation results. Although methods based on knowledge tracing can match exercises according to the level of knowledge mastery, there are still limitations: first, the modeling of the correlations among students, exercises and knowledge points is inadequate, and fine-grained answering information is neglected, which reduces the accuracy of knowledge state prediction and exercise recommendation; second, the common knowledge characteristics of student groups are not fully utilized, making it difficult to improve the novelty and diversity of recommendations. To address these problems, this paper proposes a Knowledge State Prediction and Clustering-driven Exercise Recommendation method (KSPC-ER) that integrates graph knowledge tracing with clustering technology. This method adopts dual-graph knowledge tracing to accurately depict students’ knowledge states, uses K-means++ to explore the common learning characteristics of student groups, and designs a multi-skill associated difficulty calculation mechanism and a dynamic recommendation threshold to achieve precise matching of exercises for students. Experimental results show that the proposed method outperforms the existing baseline models in recommendation accuracy, novelty and diversity, which verifies its effectiveness and superiority.