Research On Key Issues Of Multi-granularity Sentiment Analysis Of English Text

With the raise of online social networking,the Internet has gradually become an important carrier for people’s sentimental expression and opinion dissemination,which has also given birth to the development of online public opinion.How to effectively obtain people’s sentimental tendencies from Internet public opinion has become a new challenge for researchers,which is also an effective means to maintain the security and stability of the Internet.More than half of the texts on the Internet are written in English,so the research on sentiment analysis of English texts is universal and practical.We realize the innovation of sentiment analysis method from multiple granularities of English text,such as words,sentences and documents,and improves the accuracy of sentiment analysis results.We also implement a sentiment analysis system based on the research of the thesis,which can maintain cybersecurity from the perspective of sentiment analysis.We conduct a series of researches on sentiment analysis at different granularities.The main research contents are as follows:(1)Aiming at the problem that the sentiment representation information of words in the sentiment representation lexicon is not rich,we construct a Concept-based Sentiment Representation Lexicon for English Texts(CSRL).CSRL designs the sentiment representation of English text as a digital representation of multiple sentiment features,including multiple sentiments and intensities,sentiment concepts and correlations of words.First,the sentimental seed words are selected based on the six basic emotions of Ekman P.,and the multivariate sentimental intensity values of the sentimental seed words are calculated by the conversion method.Then,based on the Wordnet,the synonyms of the sentimental seed words are expanded,and the multi-sentimental intensity of the sentimental seed words is propagated to each sentimental synonym by the graph propagation method.Then,based on the sentiment words after synonym expansion and their sentiment intensity,the Hashtag Emotion Corpus and the expansion method of sentiment word based on multi-information fusion,the sentiment words are further expanded.The sentiment concept and its correlation features of each sentiment word are extracted from the Probase concept knowledge base.After the final fusion,the multiple sentiment features of the sentimental words are obtained,and the CSRL is constructed.CSRL enriches the sentiment representation information of words in sentiment representation lexicon and improves the application performance of sentiment representation lexicon.(2)In order to solve the problem of sentiment ambiguity in sentence sentiment analysis,we propose a Sentence Sentiment Analysis Model Based on Contextual Sentiment Perception(SSAM-CSP)to perform sentiment analysis on sentences.First,SSAM-CSP extracts the sentiment words in the sentence,and combines the CSRL constructed in Chapter3 and the best sentiment concept of sentiment words to obtain the sentiment concept features of the sentence,and enhances the sentiment information of words in the context.Second,SSAM-CSP learns the syntactic dependencies between sentiment words and contexts through a graph attention network,obtains the syntactic dependency features of sentences,and enhances the sentimental dependencies of sentiment words in context.Then,the sentimental concept feature and syntactic dependency feature of the sentence are interacted with the attention mechanism to obtain the sentimental concept perception feature and syntactic dependency perception feature of sentences.Finally,the two feature representations are connected to each other and processed by the decoder.SSAM-CSP effectively alleviates the problem of sentiment ambiguity in sentence sentiment analysis by enhancing the sentiment information between sentiment words and context and enhancing sentiment dependence between sentiment words and context.(3)In order to improve the accuracy of document sentiment analysis,we propose a Document Sentiment Analysis Model Based on Clause Correlation(DSAM-CC)to analysis the documents.First,DSAM-CC extracts the sentiment correlation features between clauses in the clause sentiment correlation layer,and the sentiment tendency of the previous clause is associated with the sentiment judgment of the next clause.Then,DSAM-CC extracts the structural association features between clauses based on the RST topology in the clause structure association layer,and the discourse structure of the document and the structural relations among clauses are fully considered.Based on the sentiment correlation features and structural correlation features of the clauses,the sentiment tendency of the document can be refined into the joint expression of the sentiments of each clause,and then from the level of clause correlation,combined with SSAM-CSP model in Chapter 4,the sentiment tendency analysis of the whole document can be realized.The DSAM-CC model analyzes the sentiment tendency of documents from two aspects of clause sentiment correlation and clause structure correlation,which improves the accuracy of document sentiment analysis.(4)On the basis of the above research,we implement a multi-granularity sentiment analysis experimental system for English comments,and show the interface of the system’s analysis results of English comments.