作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• •    

序列特征与学习过程融合的知识追踪模型

  • 发布日期:2023-11-29

Knowledge Tracing Model Based on the Fusion of Sequence Features and Learning Processes

  • Published:2023-11-29

摘要: 知识追踪是人工智能技术与教育相结合的新兴领域,旨在通过学生过去完成习题的交互序列对学生的知识状态进行评估,是实现大规模个性化学习服务的关键核心技术。随着深度学习在计算机视觉、自然语言处理、推荐系统等领域的广泛应用,知识追踪领域也出现了大量基于神经网络的方法,简称深度知识追踪模型。针对目前已有的深度知识追踪模型在可解释性和准确性方面的不足,本文提出了一种序列特征与学习过程融合的知识追踪模型SLKT,模型包括知识状态模块、序列特征模块、预测模块,知识状态模块用以模拟学生学习过程,序列特征模块捕捉学习者近期学习状况。通过序列特征和学习过程的融合,SLKT模型有效解决了基于知识状态建模的方法无法考虑学习者近期学习状况的问题,同时提出了一种带约束的动态Q矩阵表示练习和知识点之间的关系,从而更好地进行学习者学习过程建模,在确保较好的可解释性的同时有效提升模型的准确性。本文在三个知识追踪领域公共数据集上通过与DKT、DKVMN、SAKT、CKT等深度追踪模型进行的多组实验对比,验证了本文模型在AUC、ACC指标评估中的优越表现。

Abstract: Knowledge tracing is a new field combining artificial intelligence technology and education. It aims to evaluate the knowledge state of students through the interactive sequence of exercises completed by students in the past. It is a key core technology to realize large-scale personalized learning services. With the wide application of deep learning in computer vision, natural language processing, recommendation system and other fields, a large number of neural network-based methods, referred to as deep knowledge tracing model, have emerged in the field of knowledge tracing. Aiming at the shortcomings of the existing deep knowledge tracing models in terms of interpretability and accuracy, this paper proposes a knowledge tracing model SLKT that integrates sequence feature and learning process, including knowledge state module, sequence feature module and prediction module. The knowledge state module is used to simulate the learning process of students, and the sequence feature module captures the recent learning status of learners. Through the integration of sequence features and learning process, SLKT model effectively solves the problem that the method based on knowledge state modeling cannot consider learners' recent learning status. Meanwhile, a dynamic Q-matrix with constraints is proposed to represent the relationship between exercises and knowledge points, so as to better model learners' learning process. It can ensure better interpretability and improve the accuracy of the model effectively. In this paper, the superior performance of the proposed model in the evaluation of AUC and ACC indicators is verified by comparing multiple sets of experiments with depth tracing models such as DKT, DKVMN, SAKT and CKT on the public data sets in three knowledge tracing fields.