私播课论坛中学习者会话行为建模研究

Study of Learners'Conversational Behavior Modeling in SPOC Forums

私播课(Small Private Online Course,SPOC)论坛中的非结构性文本蕴含学习者认知和内部心理加工过程,对其分析有助于理解和解释学习结果的成因。以64名学生在SPOC论坛中的会话文本为对象进行数据挖掘,结合LDA主题建模和隐马尔可夫模型对学习者会话行为进行实体建模,并比较高低绩效组会话行为差异,最后运用回归分析和卡方检验探索了影响学习绩效的行为模式。结论表明高低绩效组的学习行为转移存在明显差异,高绩效组的行为转移具有渐进性和平滑性,倾向于序次解决问题,而低绩效组的行为转移则更倾向于浅层回溯。信息查阅、信息加工、信息发布、协作交互、问题解决和信息评价行为均与学习成绩有关,但信息查阅行为对学习成绩有着显著正向影响,且较多的协作交互和信息评价行为能够触发学习者的高阶认知。通过教育文本数据挖掘,教师能够发现不同群体的行为特征,从而进行适应性指导和精准教学,促进学习者高阶思维发展。

The analysis of unstructured text in the SPOC forum,which contains learners'cognitive and internal mental processing,is helpful to understand and explain the causes of learning outcomes.Based on the data mining of 64 students'conversational texts in the SPOC forum,this paper combines LDA topic modeling and Hidden Markov Models to model the learners'conversational behaviors,and compares the differences in conversational behaviors between the high and low performance groups,and finally explores the behavioral patterns affecting learning performance through regression analysis and chi-square tests.The results show that there are significant differences in the transfer of learning behaviors between the high and low performance groups.The behavioral transfer in the high performance group is gradual and smooth,tending towards sequential problem solving,whereas the behavioral transfer in the low performance group tends to be more shallow and retrospective.Information retrieval,information processing,information release,cooperative interaction,problem solving,and information evaluation are all related to learning performance,but information retrieval has a significant positive effect on learning achievement.And more collaborative interaction and information evaluation behaviors can trigger learners'higher-order cognition.Through educational text data mining,teachers can identify the behavioral characteristics of different groups,so as to provide adaptive guidance and precise teaching,and promote the development of learners'high-order thinking.