Text Sentiment Classification Of Hotel Field And The Research On Sentiment Element Extraction

With the popularity of the Internet and the rapid development of e-commerce,more and more people are willing to share their life experiences on the Internet and express their opinions and insights on the topics of interest to them.A large amount of textual information is generated,and in order to obtain the valuable data in the massive text,the sentiment analysis work of the text emerges at the historic moment.Sentiment analysis work can dig out the emotional expression information of users in the text,and such information can be applied to social opinion analysis,product recommendation,product research,human-computer interaction and other aspects.The main research topic of this thesis is sentiment classification,which is to judge whether the user's emotional tendency toward the research goal is a positive or negative attitude.At the same time,this thesis also judges the emotional tendency of the same emotional words to different opinion targets in a particular field,and applies it to the sentiment classification to judge the influence of fine-grained quantification of emotional words on sentiment classification.The specific work content of this thesis is as follows:(1)The principle and application methods of sentiment classification work and sentiment element extraction work are introduced in detail.It also describes the methods of selection and representation of text features,classical text classification algorithm and the flow of text preprocessing,and puts forward the difficulties in sentiment classification work at the same time,which laid a foundation for the follow-up work.(2)In order to make full use of the emotional value and semantic information in the short texts,a method of Chinese text sentiment classification combining syntactic rules,emotional value and word vectors is proposed in this thesis.Firstly,Word2vec is used to train the corpus and transform the vocabulary into semantic vector form.Then,combining the syntactic rules and the word vector,the vector representation of the comment text is obtained.Because of the bad result by universal sentiment lexicon has on the sentiment classification in particular fields,a sentiment lexicon creation method in particular fields is proposed in this thesis.The method can create an sentiment lexicon that fits to the text sentiment classification in the field of hotel.By combining the sentiment lexicon and the syntactic rules,emotional values of each text can be obtained,and then a sentiment model Vector with Emotional Value(VWEV)which combining the short text vector and the emotional value is constructed to classify text sentiment.Finally,this thesis uses the Support Vector Machine(SVM)algorithm to construct the classifier model and obtains the optimal feature selection method and the optimal number of features through experiments.(3)The emotional tendency of the same emotional word to different opinion targets may be different.Therefore,this thesis attempts to find out the emotional tendencies of emotional words for different opinion targets,and uses the information to re-classify the text.In this thesis,sentiment elements are extracted jointly based on the syntactic rule template.Meanwhile,cosine similarity of word vectors is calculated by using Word2vec.The features of synonymous opinion targets are clustered and the names of opinion targets are unified.Then,the frequency of joint occurrence of each opinion target and each opinion word in positive or negative texts is recorded,and the emotional tendency of each opinion word to each opinion target is obtained by using it,and then the joint sentiment lexicon is generated by combining the constructed sentiment lexicon.Finally,the sentiment classification experiment is re-examined on the original data set with the lexicon.It is found that the method improves the results of sentiment classification and verifies the effectiveness of the method.