【摘要】 教学过程中产生的群聊文本往往包含着学生对于课程的思考。通过提取分析群聊文本中学生的提问,能够了解学生的学习情况,也能结合具体内容对学生进行指导。该研究旨在通过构建问句抽取模型,对群聊文本中与课程相关的提问进行提取。实验首先针对教学过程中产生的群聊文本进行收集,并结合课程相关教材进行数据清洗工作;然后针对类别分布不均匀问题,在Text-CNN模型的基础上提出了两种优化方式:引入注意力机制和使用平衡交叉熵损失函数。实验结果表明,优化后的模型能够达到91.95%的正确率,比原有模型增加了1.04%,而问句的F1-score表现为0.72,在原有模型的基础上提高了0.06。该模型能够运用到实际教学中,将群聊文本中与教学相关的学生提问抽取出来,再与线下教学相结合,提高教师的分析效率,进一步改善教学效果。更多还原

【Abstract】 The group chat text generated in the teaching process generally incorporates the thoughts of students regarding the course, thus, extract and analyze students’ questions in the group chat text can help teacher get a better understanding of the students’ learning progress and guide students according to the learning content. This article aims to extract questions related to the course in the group chat text by constructing a question extraction model. Specifically, the group chat content generated during the teaching process is collected in the first step, and then, by employing course-related materials, the data cleaning is conducted.To tackle the problem of uneven categories, two optimized methods are proposed based on the Text-CNN model, i.e., the attention mechanism and the balanced cross-entropy loss function. The experimental results show that the optimized model achieves the correct rate of 91.95 percent which increases 1.04 percent than the original model. Meanwhile, the F1-score of the question sentence is 0.72, which is higher than the original one by 0.06. The results provide the evidence that this model can be applied to the actual teaching to extract those questions related to the teaching process from the group chat text, to combine the offline teaching process, to improve the efficiency of the teachers’ analysis work, and further to improve the learning effect.更多还原