Chinese Text Sentiment Orientation Analysis For MOOC Online Comments

With the rise of the concept “Internet + education”,Massive Open Online Course(MOOC)has developed unprecedentedly in these years.As we can see,more and more users begin to use the MOOC platform to select courses which they are interested in,and leave lots of comments that contain personal emotions.The number of comments has exploded,and these comments always contain the information about learners’ attitudes and evaluations of courses.It is very important for learners,teachers,and the administrators of MOOC platforms to analyze and process these commentary texts so as to obtain critical information.As a key step of text analysis,sentiment orientation analysis has become a hot research topic naturally.At present,there are two main methods in the task of text sentiment orientation analysis: unsupervised and supervised methods.In this paper,based on the Chinese commentary texts of MOOC,aiming at the problems of traditional classification methods and taking full account of the hierarchical structure of Chinese texts and the characteristics such as semantic information with single character,an unsupervised sentiment analysis method and a supervised sentiment orientation analysis method are proposed respectively.The mainstream unsupervised method is the method based on sentiment lexicon.When this method is used in the analysis of MOOC comments,there are many problems such as the coverage of sentiment words is not wide,the migration of the field is weak,and the calculation method of sentiment value is not comprehensive.Aiming at these problems,this paper proposes an unsupervised text sentiment orientation analysis method based on the sentiment lexicon in the field of MOOC.Firstly,the three sentiment lexicons published online are merged and de-duplicated by us.Then,the sentiment lexicon in the field of MOOC courses is expanded and formed by using Word2 vec tool.Finally,the sentiment values of the commentary texts are obtained by matching the new sentiment lexicon by introducing degree adverbs,negatives and nouns into the calculation formula.Among the supervised methods,deep learning method is the hottest one.However,such methods often could not take the hierarchical structure of texts into account,and could not solve the ambiguity problems of the Chinese texts very well.Aiming at these problems,this paper proposes a Convolutional Neural Network-based and Hierarchical Attention Network-based Chinese Sentiment Classification Model(C-HAN Model),which integrates convolutional neural network and hierarchical attention network.Firstly,convolutional neural network is used to capture the relationship between Chinese characters,then hierarchical attention network is used to find the key information that has the greatest impact on sentiment orientation.Finally,the best sentiment orientation analysis model is obtained by the parameter adjustment in training,in order to achieve accurate classification of sentiment orientation of commentary texts.The two methods are experimented on commentary corpus of MOOC courses,and compared with the previous methods,both of them have improved the classification effects.Especially,our experiments show that compared with the traditional sentiment lexicon,the proposed sentiment lexicon covers more sentiment words in the field of MOOC courses,and considers more influencing factors such as special nouns when calculating sentiment value,which improves the sentimental classification index F1 by about 5%.At the same time,the deep learning model C-HAN proposed in this paper combines convolutional neural network,recurrent neural network and attention model,so that it can not only solve the ambiguity problem caused by Chinese text segmentation,but also find the key information which has the greatest impact on the result of text sentiment orientation.By comparing C-HAN model with several deep learning models,it is found that our model C-HAN could achieve the best sentiment classification effect on this task,which verifies the superiority of the deep learning technology in the text analysis of MOOC comments.