Research And Application Of Sentiment Analysis Technology Based On Texts Of Teaching Evaluation

Ministry of Education promulgated the first national standard for teaching quality in the field of higher education in 2018,which reflects the great emphasis on improving the quality of teaching.Students are the receivers of teaching,and they have intuitive and clear feelings about the quality.Mining evaluations can effectively reflect the quality of the course.Educational administration system stores a large number of unstructured evaluation texts,which is difficult to make quantitative and qualitative analysis.This paper uses natural language processing techniques and machine learning methods to conduct coarse-grained emotional scoring and fine-grained topic mining for teaching evaluation,aiming at helping school managers improve the quality of teaching.Main research results are as follows:1.A framework for automatic construction of domain sentiment lexicon is proposed.Based on the framework,we constructs a Weighted-Edu-Dictionary,which is used to evaluate the coarse-grained emotions of teaching evaluation.The process of constructing sentiment lexicon: Firstly,we present an automated approach to select seed words from basic knowledge and domain corpus.This method can identify seed words with stable emotional tendency,high emotional intensity and high frequency.Then,we use PMI to calculate the co-occurrence degree of the seed word and the candidate word,which is used as the weight of the association graph.Finally,the label propagation algorithm is used to mark the polarity of the candidate words.Line standardized the final tag matrix.The value in label matrix is used as the weight of the emotional word.PMI can measure the similarity between words,but it ignores the complex sentences.This paper introduces the idea of emotional annotation of CRM algorithm into the calculation of PMI,which makes the PMI calculation result more reliable.The quantitative calculation method of emotional scores uses dependency syntax to identify the four characteristics of the emotional scores in the evaluation text.According to the influence of each feature on the emotions,a method for calculating the emotional scores is defined.We can use the method to calculate the positive and negative sentiment score of the given comments.The construction framework proposed can be used to construct emotional dictionary in other fields.The constructed Weighted-Edu-Dictionary dictionary enriches the Chinese sentiment dictionary.The calculated emotional score can be used to analysis the quality of teaching in different dimensions such as curriculum,teachers and colleges.2.A fine-grained topic mining model for teaching texts is proposed.The model extracts the explicit topics in the evaluation based on the dependency syntax.We propose a calculationmethod for the similarity of the topics.The method is based on the similarity of the subject’s own structure and the similarity between the corresponding emotional words of the topic.Combine the synonymous topics based on the calculation results.According to two different types of implicit topics in two texts,we classify and extract them.The first implicit topic is modified by the exclusive emotional words.This paper identifies the implicit topic by constructing a co-occurrence matrix that displays the topic and the emotional words.The second type of implicit topic is modified by special emotional words.The emotional words do not appear together with the topic in the corpus,but the emotional words themselves imply an attitude towards a certain topic.This paper constructs <topic,special emotional words> table artificially to identify such implicit topics.The fine-grained topic mining model constructed in this paper can extract the topics in the evaluation text more accurately.3.A web-based evaluation text mining system with research results is designed and implemented.The main functions include: teaching evaluation and information maintenance of related data,calculating the positive and negative scores based on the Weighted-Edu-Dictionary dictionary,and identifying the tendency of the subject to be extremely emotionally inclined in the evaluation text.In addition,users can analyze the association between research result and basic information according to different dimensions such as courses,departments,teachers,etc.The system visualizes the result data through tables,histograms,and line charts.