Multi-grain Sentiment Analysis Of Teaching Reviews Based On Topic

The “Double First-Class” initiative,another major national strategy for higher education newly launched by China,aims to make the quality of teaching a top priority for the long-term development of universities.Student's evaluation of teaching is an effective way of assessing teaching quality.So how to quantify and use the large amount of textual information in the assessment system holds the key to evaluating teaching quality in a fair and objective manner.This paper focuses on sentiment analysis of teaching reviews through natural language processing and machine learning methods,and aims to objectively and impartially feedback the relevant information of teachers' teaching quality.The main research work is as follows:1.Compiled Topic-Dictionary,a topic dictionary for analyzing teaching reviews.This paper proposes an extraction algorithm based on the principle of co-occurrence of words.The algorithm extracts a collection of topic words and emotion words,and provides a new effective method for the noise elimination of data.As a result,the topic collection boils down to nine major categories,and with it comes the topic dictionary.2.Compiled Edu-Dictionary,a sentiment dictionary for analyzing teaching reviews.This thesis uses Word2 Vec method to expand the dictionary,a method based on word vectors carrying a wealth of semantic information.This potential semantic relationship between word vectors helps effectively expand the dictionary.So a special sentiment dictionary which is built on the corpus of teaching evaluation can effectively identify new words and unique expressions in the field of teaching evaluation.3.Proposed a fine-grained sentiment analysis model based on topic.This model,which bases on the analysis of Chinese syntactic dependency and combines with topic dictionaries,sentiment lexicons,degree adverbs dictionaries and negative lexicons,constructs a “theme,emotions” evaluation unit,calculates the emotional polarity and obtains the final assessment.4.Proposed a coarse-grained sentiment analysis model based on sentiment dictionary and machine learning.Firstly,the sentiment dictionary helps to calculate the emotional polarity,followed by the evaluation text with higher confidence forming a training data set to train the classifier.Then SVM_KNN classifier with multi-feature fusion processes a test data set and experiments show that SVM_KNN classifier outperforms a single classifier.