Research On Sentiment Analysis Of Student Evaluations Of Teaching Based On Deep Learning

In October 2020,the CPC Central Committee and State Council issued the “Overall Plan for Deepening Educational Evaluation Reform in the New Era”,which embodies the importance of improving teaching quality in the strategy of revitalizing China through education.Students are the target audience of education and have a clear sense of the quality of teaching.Excavating the emotional tendency in student evaluations of teaching can reflect the real teaching situation and help improve teaching.To this end,in response to the current major problems in the field of student evaluations of teaching,this thesis uses deep learning methods to conduct sentiment analysis research on the texts of student evaluations of teaching,and proposes two deep learning models for aspect-level sentiment analysis and cross-domain sentiment analysis of student evaluations of teaching,details as follows:In order to accurately find out the emotional tendency of student evaluations of teaching and help improve teaching,a deep memory network model with prior information is proposed.Firstly,the comments are vectorized by word embedding.Second,the bidirectional gate recurrent unit is used to model the comments to obtain their semantic information,which is used as the memory module of the deep memory network.Then,other information outside the comments is used as prior information,and extract related emotional features from the memory module through the attention mechanism.Finally,use the gated recurrent unit to update it,so as to obtain accurate emotional classification.Experiments on the data set of student evaluations of teaching show that the proposed method can effectively explore the emotional tendencies of different teaching aspects in the comments,and provide an effective way for teachers and teaching administrators to understand and improve teaching.In order to solve the problem of the scarcity of labeled data in the field of student evaluations of teaching,a cross-domain sentiment analysis model based on aspect interactive transfer network is proposed.Firstly,model comments and aspects through word embedding and bidirectional gate recurrent unit separately to obtain their semantic representations.Then through the aspect interaction module to realize the interaction of the comments and aspects,and use the deep memory network and attention mechanism to fully mining the hidden features of domains and emotional in the comments.Finally,training domain classifiers and sentiment classifiers through adversarial training,so that the model can learn the common emotional features between the source domain and the target domain to realize cross-domain sentiment analysis.Experiments on data sets of product reviews and student evaluations of teaching show that the model proposed in this thesis has better classification performance in cross-domain classification,and can effectively find out the emotional tendency of the comments in student evaluations of teaching.