摘要
为及时准确地对学生成绩进行预测,提出基于多层特征融合的学生成绩预测模型。针对以往研究未能对成绩信息进行有效特征表示,导致预测效果不佳的问题,利用LSTM和注意力机制的柔性结合方式实现成绩序列信息在课程和时间两个维度的同步特征提取。针对学业早期训练数据不足问题,构建基于时间共现频率的相似学生计算方法,融合相似学生信息实现信息互补。实验结果表明,该模型的准确度、稳定性和及时性都高于其它基线方法。
A student grade prediction model based on multilayer feature fusion was proposed for timely and accurate prediction of student grades.To address the problem of poor prediction due to the failure of previous studies to effectively represent the grade information with features,a flexible combination of LSTM and attention mechanism was used to achieve simultaneous feature extraction of grade sequence information in both course and time dimensions.To address the problem of insufficient academic early training data,a similar student calculation method based on temporal co-occurrence frequency was constructed to fuse similar student information to achieve information complementarity.Experimental results show that the accuracy,stability and timeliness of the model are higher than those of other baseline methods.
作者
刘彤
齐慧冉
倪维健
LIU Tong;QI Hui-ran;NI Wei-jian(College of Computer Science and Engineering,Shandong University of Science and Technology,Qingdao 266590,China)
出处
《计算机工程与设计》
北大核心
2023年第10期2973-2978,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(71704096)
山东省自然科学基金项目(ZR2022MF319)
青岛社会科学规划基金项目(QDSKL1801122、QDSKL2001117)
山东科技大学青年教师教学拔尖人才培养基金项目(BJ20211110)
山东科技大学教育教学研究“群星计划”基金项目(QX2020M25)。
关键词
学生成绩预测
特征融合
长短期记忆网络
注意力机制
历史成绩建模
共现频率
特征提取
student grade prediction
feature fusion
long-short term memory
attention mechanism
historical achievement modeling
co-occurrence frequency
feature extraction
