【作 者】：
熊余，张健，王盈，蔡婷
【关 键 词 】：
知识追踪； 深度学习； 教育大数据； 认知评估； 深度兴趣演化网络
【栏 目】：
理论探讨
【中文摘要】：
知识追踪旨在根据学生历史学习记录实时追踪学生的知识水平变化，以预测学生在目标试题上的表现，从而推动学生知识状态自动评估、学习策略个性规划、试题资源精准推荐等智慧教育愿景的实现。为了准确挖掘学生潜在的知识掌握情况，并考虑试题顺序对预测效果的影响，文章提出一种基于深度学习的演化知识追踪模型（简称深度演化知识追踪模型）。该模型通过引入深度兴趣演化网络来动态感知学生知识水平变化，推断学生的知识状态及演化规律，从而预测学生在目标试题上的表现。实验证明，该模型有效地提高了未来试题表现的预测准确度，在精准挖掘学生知识状态的同时，能够动态感知其知识的演化轨迹。
【英文摘要】：
Knowledge tracking aims to track the changes of students' knowledge level in real time according to their historical learning records and predict students' performance on target questions, so as to promote the realization of smart education visions such as automatic assessment of students' knowledge status, personalized planning of learning strategies and accurate recommendations of test resources. In order to accurately explore students' potential knowledge mastery and consider the influence of question order on the prediction effect, an evolutionary knowledge tracking model based on deep learning (i.e. Deep Evolution Knowledge Tracing, DEKT) is proposed. This model can dynamically perceive the changes of students' knowledge status by introducing a deep interest evolutionary network, infers students' knowledge status and evolutionary patterns, and thus predicts students' performance on target questions. Experimental results show that this model effectively improves the prediction accuracy of future test performance, and can dynamically perceive the evolutionary trajectory of students' knowledge while accurately mining their knowledge status.