Research On The Evaluation Model Of Learner's Learning Input In MOOC Discussion Area

The emergence of online learning platforms has promoted the development of education modernization.MOOC is one of the most representative online learning platforms.Its forum is a place for learners to have interactive communication.Here,learners can discuss about learning doubts,experiences or difficulties.Learning engagement is the premise of effective learning,which mainly includes behavioral,cognitive and emotional engagement.The degree of learning engagement shows the extent of the learners' engagement in the learning process and reflects the level of learning engagement that the learners can achieve.The study of learning engagement is mostly based on the classroom,and there is relatively little research on online learning engagement.Secondly,the researches focus more on the "quantity" of learners' behavioral engagement,and less on emotional engagement.There is even less research in comprehensive consideration of behavioral,cognitive and emotional aspects of learning engagement.Based on the above background,this study took a course on the MOOC platform as an example to explore the influencing factors and evaluation methods of online learning engagement from the three dimensions of behavior,cognition and emotion,in order to evaluate online learning more scientifically and improve the teaching and learning of the online courses.Based on the existing research,this paper puts forward a learning engagement evaluation model of MOOC forum,and introduces 16 characteristic indexes in detail from three dimensions of behavior,cognition and emotion.According to the existing research and the data characteristics of this study,the acquired data was divided into16 related features according to behavior,cognition and emotion.Among them,there are 9 behavioral characteristics: the number of questions answered,the number of views,the number of votes,the number of participants in the topic,the duration of the topic,the distance between the question and final exam,the level of interaction,the type of discussion forum,whether the question is anonymous;5 cognitive characteristics: the type of the question,whether the question is expressed clearly or not,the type of knowledge,the importance of knowledge points,whether the teacher participates or not;2 emotional characteristics: the questioner's emotional tendency and the average emotional tendency in the response.The thesis was based on a national excellent course named "Python Language Programming" with a large number of participants on the MOOC platform(accumulated learners exceeding 1.73 million)as a sample to verify and implement the model.The sampling time was from March 13,2018 to July 10,2018.Python was used to write a crawler to crawl the interactive data in the discussion forums,including 14 related contents,such as question ID,question topic,question description,and picture links included in the question,etc.It was cleaned and filtered according to time and content,and finally retained 20,434 data and 6,718 topics.According to the research,we found that: 1.Model Indicators.Among the behavioral characteristics,"anonymity or not" had no significant effect on learning engagement(P> 0.05),and the other 8 variables had significant effects on learning engagement(P <0.001).Among the cognitive characteristics,5 variables had significant effects on the degree of learning input(P <0.001).However,due to the high collinearity of problem types and knowledge types,only problem types were retained.Among the emotional characteristics,2 variables had significant influence on learning engagement(P <0.001).Based on the analysis of specific cases,the evaluation model is adjusted and 14 evaluation indexes are retained from three dimensions of behavior,cognition and emotion.2.Model algorithm implementation.The logistic regression,SVM,decision tree and random forest were used to construct the binary classification model.Due to the imbalance of data caused by the deep engagement accounting for only 2.86% of the total topics,the middle and deep engagement are combined as the deep engagement to distinguish it from the shallow engagement.First,use the full set to adjust the parameters of the four algorithms to find the optimal parameter combination;Then ranking according to the related to learning engagement,the top 6(subset 1),the top 8(subset 2),the top 10(subset 3)and all variables(full set)were selected as feature sets respectively.They were trained in four classifiers to find the optimal feature subset.The results showed that the highest precision of the four algorithms is more than 80%.Among them,the performance of logical regression on subset 3 is better,with an precision of 80.18%;the performance of decision tree on subset 2 is better,with an precision of 81.80%;the performance of SVM and random forest on the whole set is better,with an precision of 82.63% and 82.34%.The SVM classifier can be trained by full set in the following modeling optimization for future.