Research On Thesis Review Data Analysis And Evaluation System Based On Sentiment Classification

With the continuous expansion of the scale of postgraduate education in China,its development direction has shifted from the initial scale expansion based on enrollment growth to the connotative development based on training quality.The expert review of dissertations is a comprehensive and specific evaluation of the quality of students ’dissertations by experts.The analysis of the expert review of dissertations is helpful to improve the quality of dissertations.However,in the face of a large number of unstructured review text data,manual screening is inefficient and resource-consuming.In view of the above problems,this thesis studies the existing sentiment classification methods,designs and implements an external review text analysis and evaluation system based on sentiment classification,and provides decision support for relevant management departments.The main research contents of this thesis are as follows :(1)Data acquisition and preprocessing.First of all,this thesis uses the crawler to crawl the thesis review data from the Open Review platform and translate it into Chinese.It deletes and annotates the garbled,semantically unsmooth and invalid information after text translation,and constructs the thesis review data set.Finally,BERT-Muti CNN,BERT-LSTM and BERT-GRU are used to carry out five-fold cross validation experiments.The experimental results show that the thesis review data set constructed in this thesis has good robustness.(2)BERT-MSCNN model construction.Aiming at the task of external review sentiment classification,a BERT-MSCNN sentiment classification model based on multi-scale convolutional neural network and squeeze-excitation network is designed on the basis of BERT pre-training model.Firstly,for the problem that Word2 Vec model cannot distinguish the meaning of words in different contexts,BERT dynamic pre-training model is used to vectorize the text.Secondly,multi-scale convolutional neural network is used to extract local semantic features.Finally,the squeeze-excitation network SENet is introduced to calibrate the weights of the feature channels,strengthen the extraction of features,and improve the recognition effect of the model.The experimental results show that the proposed model can effectively classify the sentiment classification of the review data of the thesis,and its accuracy can reach 78.12 %.Compared with the sentiment classification model of multiple fusion self-attention mechanisms,it has higher recognition rate.(3)Thesis review emotion classification system development.Based on the constructed BERT-MSCNN deep learning model,this thesis uses the idea of front-end separation to construct the system,and uses Vue.js,Echarts and Flask frameworks to develop a thesis review data analysis and evaluation system based on sentiment classification.The functions of data import,emotion analysis,comment retrieval and word cloud display are realized,and finally the function test and model validity test are passed.