Research On Aspect Based Sentiment Analysis Of MOOC Comments

The rapid development of the Internet has provided people with opportunities for online learning.Many people who want to improve their knowledge and professional skills would like to participate in online education.A large amount of interaction data has been generated through the learning process.These data truly reflect learners' interests,learning experience,emotional attitudes and so on.Therefore,the analysis results of educational interactive data could help learners find learning resources,help teachers organize teaching content,and help the platform strengthen the infrastructure.The MOOC platform of China university is one of the largest platforms which provide online courses.This paper uses crawler to obtain course comments of students on MOOC and designs a model named Graph Convolutional Network for Aspect Based Sentiment Analysis,shortened to GCN?ABSA,to analyze the aspect based sentiment.The main work is concluded as follows:(1)Extracting aspects of comments is actually a named entity recognition in natural language processing,whether the word embedding learned from the text is reasonable or not is the key to entity recognition.Dynamic word embedding can learn semantic relationship between words.In this paper,we will use Bert pre?trained model to initialize text vector representation,and then adjust the module parameters through training on a smaller data set,for a better effect on our research field.However,when labeling named entities,we explore the optimal annotation sequence on conditional random field from a global perspective.(2)Emotional classification in aspect based sentiment analysis,to some extent,determines its emotional category of the labeled aspects.Different from all kinds of deep learning models,This paper uses the rules of emotional characteristics learned from corpus to identify the emotional categories of aspects.We will integrate the syntactic information into the deep learning model through graph convolution neural network,and then established the relationship between aspects and emotional classification to make the deep learning model more interpretable.we also design a context masking layer in our model,which sets the weights of other tokens according to the distance from aspects,The encoding process is carried out through a multi-head attention mechanism.Finally,the hidden state of the first position is obtained,and the emotion is divided by a softmax function.(3)We collect course comments from MOOC to evaluate the GCN?ABSA model.at the same time,we use unsupervised learning methods to mine hidden topics in comments.The aspects extracted by GCN?ABSA model are used as emotional topic groups.Finally,the mapping from ”text-aspects-emotional topics” to ”emotional topics-themes ” would be established.