Research On Negative Comments Of Online Courses For Unbalanced Data

With the cross-border integration of "Internet +" and various industries,a large number of "Internet + Education" online course platforms have emerged,such as China University MOOC,Netease Cloud Classroom and Tencent Class,etc.These platforms provide a large number of efficient and practical courses,which have attracted the participation of many students.However,the platforms also have problems such as uneven teaching quality and incomplete teaching functions.The course comments are a comprehensive evaluation of the course by students and reflect their emotional attitude towards the courses.Therefore,mining and analyzing course comments is helpful to analyze and evaluate the quality of the courses as well as the construction and improvement of online education platforms.Negative comments in course reviews are relatively small,but are particularly important.They reflect the problems encountered by students in the learning process and their opinions on courses,teachers and platforms.These negative comments have important reference value in helping students choose courses,improving teaching methods,and content and platform construction.Based on this,this paper focuses on the negative comments of the courses on the online education platforms.By categorizing the positive and negative tendencies of course comments on online education platforms,identifies negative comments among them,and conducts topic mining on negative comments,which can provide suggestions for teaching content,teaching quality and platform construction to increase the participation rate of the courses and students’ learning motivation.Because there are far more positive comments than negative comments in course comments,that is,course comments are highly unbalanced.If traditional classifiers are directly used to classify course comments,the classification effect will be difficult to guarantee.Therefore,how to analyze and mine this highly unbalanced dataset is also the focus of this paper.The main work of this paper includes:(1)we cleaned the English online course comments collected in China University MOOC and Tencent Class,and obtained 21,243 pieces of comments,among which the ratio of positive and negative comments reached 19: 1,which is a highly unbalanced dataset.(2)Use oversampling techniques to sample the data,such as SMOTE,Borderline-SMOTE1,Borderline-SMOTE2 and ADASYN,and then in combination with traditional machine learning classification models to classify the course comments,such as Support Vector Machine,Na?ve Bayes,Decision Tree and Random Forest.At the same time,the cost sensitive focus loss function is used to improve the convolutional neural network to classify course comments.(3)In order to extract topic features of negative comments of online course comments,use Non-negative Matrix Factorization to model the negative comments after the model classification.(4)This article also extracts the negative comments in the summary articles of online teaching,analyzes and mines the shortcomings of online courses from the perspective of teachers,and makes the research on negative comments of online courses more comprehensive and sufficient.The research results of this paper are reflected in unbalanced data processing,negative comment topic mining and suggestions for English online courses.First,the classification effect of the improved convolutional neural network based on focus loss function is superior to the traditional machine learning classification algorithm based on oversampling technology in this extremely unbalanced text dataset.Secondly,topic modeling for negative comments can more effectively dig out the shortcomings of the course than topic modeling directly for comments without the classification of positive and negative tendencies.Finally,the main shortcomings of English online courses are as follows: multimedia teaching function are not perfect,the network of online education platforms are unstable,the online education platforms do not provide subtitles or the subtitles do not match the sound,teachers’ spoken language are not authentic,teacher lacks feedback on the student’s learning status,students are less involved in online teaching interactions,etc.these reflect the improvement of English online courses.