Research On Sentiment Classification For Teaching Evaluation

The task of text sentiment classification is to process and analyze sentimental texts,and summarize them into categories.Teaching evaluation texts are students’ specific evaluations of teachers’ teaching effects and are an important means of education quality monitoring.However,manual identification methods are inefficient for numerous and complicated teaching evaluationt texts.Therefore,efficient sentiment analysis technology is urgently needed to extract effective information of students’ evaluation of courses in the teaching evaluation texts.This paper proposes a sentiment classification model BERT＿LK based on BERT and Language Knowledge to complete the sentiment classification of teaching evaluation texts.The language knowledge in the model includes the syntactic relationship and the teaching evaluation texts vocabulary.The specific contents include:(1)Constructs the data set of 13,994 Chinese teaching evaluation texts.The data set is divided into three categories according to sentimental tendencies: positive,negative,and neutral.(2)Propose the BLAS model.This model combines Bi LSTM with attention mechanism and syntactic relationship.Bi LSTM can obtain the semantic information of the sentence in both directions and better learn the text features.The attention mechanism can assign different weights to words and pay attention to the words with obvious sentimental features in the sentence.The syntactic relationship solves the problem that modifier words and modified words in the sentence cannot correspond accurately.(3)Propose the BERT＿LK model.This model is improved on the basis of the BLAS model.It combines BERT and attention mechanism,and fuses the language knowledge of syntactic relationship and teaching evaluation texts vocabulary.BERT improves the accuracy of word vectors in specific contexts.The teaching evaluation texts vocabulary can more fully learn the features of the teaching evaluation texts,and highlight the role of the words containing the sentimental tendency and the evaluation theme in the texts.In order to verify the classification effect of the model,this paper uses other sentiment classification models for ablation experiments.Comparing the experimental results,this paper uses the data of data enhancement to train the model can improve the classification effect of the model.On the data set constructed in this paper,the Macro Avg＿F1 value and Weighted average＿F1 value of the BLAS model are 79.84% and 87.51% respectively.the Macro Avg＿F1 value and Weighted average＿F1 value of the BERT＿LK model are 81.38% and 88.80%respectively,which are 1.54% and 1.29% higher than BLAS,and are higher than other comparison models in this paper.Therefore,the BERT＿LK model proposed in this paper can automatically complete the sentiment classification task according to the features of the teaching evaluation texts.