Research For Sentiment Analysis And Topic Clustering Methods On Teaching Evaluation

Teaching evaluation is an important link for universities to ensure the quality of teaching.Through comprehensive analysis of teaching evaluation data,on the one hand,it can be concluded that students are satisfied with their teachers,on the other hand,it can also provide effective opinions and suggestions for the subsequent teaching work.Most of the existing utilization of teaching evaluation data is simple statistical analysis of teaching evaluation scores,while less attention is paid to teaching evaluation sentences.This thesis mainly conducts sentiment analysis and topic clustering research on teaching evaluation sentences in teaching evaluation data.The research results can help to analyze the satisfaction of students with teachers and extract teaching evaluation sentences with “suggestions”.At the same time,it is also possible to group teaching comments containing the same subject into one category to achieve more targeted analysis.At present,most of the research methods for this problem are supervised learning methods,which rely on a large number of labeled training samples.How to perform sentiment analysis and topic extraction without pre-labeling the teaching evaluation sentence and ensure reliable accuracy is a research problem.In response to the above problems,the main research contents and innovations of this thesis are as follows:Using the sentiment dictionary as a priori knowledge,combined with the recalculated teaching evaluation scores,a set of strong rules that can generate labeled training samples is developed,and then the number and characteristics of the training samples are expanded through data enhancement methods.Text CNN serves as a classification model and optimizes input vectors to predict the emotional polarity of data that has not been filtered by strong rules.Experiments show that the comprehensive accuracy of this method reaches 86.06%,which is better than other“unsupervised” methods.The defects of the EDA method for text data enhancement are analyzed,and the information gain constraint is introduced on the basis of the original method.Word-EDA method based on words and Sen-EDA method based on sentences are proposed.Experiments show that Word-EDA has improved 1.05%?2.36% on different models compared to the original method.Sen-EDA method can improve the accuracy of the model by 2.8%?3.4%.In view of the sparse text features of the teaching evaluation sentences,the trained Word2 vec word vectors are used to expand their features to increase the co-occurrence frequency of feature words of the same topic.LDA is used for topic analysis and KMeans for topic clustering.Experimental results show that the improved method improves the clustering effect by 8.1%.