毕 耕 朱晓敬 李盼池
(东北石油大学 计算机与信息技术学院，黑龙江 大庆 163318)
【摘 要】针对高校在线课程学生成绩预测问题，本文提出了基于长短期记忆（LSTM）网络的预测方法。在综合考察学习成绩影响因素的基础上，结合教育学相关理论，构建了评价输入指标集，具体包括观视比、出勤情况和课堂表现。预测结果为期末笔试成绩、期末实验成绩和阶段测试成绩。教师根据在线教学实际情况，将每个学生的输入指标以学时为单位量化成具体数值作为LSTM的输入，将真实的阶段测试、期末笔试和实验成绩作为LSTM的输出，构造训练样本集实施网络训练。训练后的网络即可用于相同课程后续学生的成绩预测。实验结果表明，预测集样本三个输出指标预测结果的均方误差均在10以内，这表明应用深度学习的预测模型解决教育学领域的学生成绩预测问题的研究思路是可行的。
【关键词】在线学习；成绩预测；LSTM
Student Achievement Prediction for Online Courses Based on Long Short-Term Memory Network
BI Geng，ZHU Xiaojing and LI Panchi
(School of Computer and Information, Northeast Petroleum University, Daqing 163318, China)
Abstract: Aiming at student achievement prediction in online courses in colleges and universities, this article proposes a prediction method based on long short-term memory (LSTM) network. After comprehensively examining influencing factors of learning achievement and combining relevant theories of pedagogy, a set of evaluation input indicators is constructed, including: observation-to-view ratio, attendance, and classroom performance. The prediction results are the final written test score, the final experimental score and the stage test score. According to the actual situation of online teaching, the teacher quantifies the input indicators of each student into specific values in units of credit hours as the input of LSTM, and takes the real stage test, final written test and experimental results as the output of LSTM, and constructs a training sample set to implement network training. The post-training network can be used to predict achierement of subsequent students in the same course. Experimental results show that the mean square error of the prediction results of the three output indicators of the prediction set sample is within 10, which shows the prediction model of deep learning, feasibility to solve the problem of student achievement prediction in the field of education.
Keywords: online learning; achierement prediction; LSTM
下载： 
基于长短期记忆网络的在线课程学生成绩预测.pdf
