Research On Sentiment Analysis Of MOOC Course Evaluations In Colleges And Universities Based On LSTM And LDA

With the rapid development of Internet and online education,online education has attracted more and more students,which not only provides different ways of online teaching mode,but also breaks the time and space limitation.MOOC is one of the popular online education platforms nowadays,and has a large number of excellent courses for learners to learn,while some learners will comment on the courses they have learned in the message area.We can get some valuable information by using sentiment analysis technology to analyze the course evaluation effectively.In this dissertation,we take the evaluation of college and university courses in MOOC as the research object,and conduct the research of sentiment polarity analysis and sentiment topic analysis on the research object.The main work of this dissertation and the innovation points are as follows:First,A multi-channel CNN-BiLSTM model based on focal loss is proposed to analyze sentiment polarity for the course evaluation.Since the dataset of the sentiment polarity classification experiment is unbalanced between positive and negative samples,we choose the focal loss function as the loss function of the model to control the positive and negative samples.The advantage of BiLSTM model is that it can effectively obtain contextual information,but its disadvantage is that it cannot extract local information,so this dissertation uses the CNN model to extract this part of information,which better overcomes the shortcomings of the BiLSTM model.Through experiments,the results prove that the method outperforms the other deep learning methods in the experiment for sentiment polarity classification.Second,a sentiment topic analysis method based on combining TF-IDF and LDA topic model is adopted to perform topic mining on course evaluations and further analyze finer-grained information.Since the traditional LDA topic model uses the bag-of-words model for text vectorization,and the bag-of-words model has shortcomings,such as the word frequency of common words is often high and the word frequency of proper nouns is low,we introduce TF-IDF in order to solve this problem.Through experiments,the results prove the effectiveness of the method.Third,using the above two models to analyze the evaluation of college and university courses in MOOC by example,identify learners’ sentiment tendency and dig out the hot issues of learners’ concern,so as to help other learners analyze and judge the quality of the course to be chosen,and help educators understand learners’ feedback on the course and further improve and refine the course.