Research On Text Sentiment Analysis And Application Based On Depth Semantic Information

The survey on satisfaction of the automotive customer can discover the potential needs of customers for products and help the enterprises make the right decisions.Now there are a large number of reviews published by users on the network,which contain the user experience and feelings about cars.Excavation and sentiment analysis of these massive online reviews can provide theoretical basis and data support for the survey on satisfaction of the automotive customer.However,due to the complicated and abstract characteristics of the text sentiment,especially the irregular of online text expression and the interference of a large amount of noise,sentiment analysis still faces many problems,such as the poor coverage of the sentiment lexicon,the difficulty of extracting sentiment features and the low accuracy of sentiment analysis.To solve the above problems,this paper takes online reviews of product as the research object,and studies construction of domain-specific sentiment lexicon based on corpus,word embedding model that integrates sentiment information and semantic information,and text sentiment analysis method based on sentiment word vectors and long short-term memory network.Finally,the above models and methods are applied to the automotive field,and an automobile customer satisfaction system based on sentiment analysis is designed and implemented.The main research work is as follows:(1)Research on the construction of domain-specific sentiment lexicon based on point mutual information and label propagation algorithm.For the problem that the traditional point mutual information method does not take corpus syntactic information into account,which causes the error in measuring sentiment correlation value between words,on the basis of the co-occurrence of words,a method of point mutual information fusion of global information,local information and constraint information is proposed to compute the sentiment correlation value of sentiment words more precisely.And the sentiment correlation graph is constructed based on the sentiment correlation value.Aiming at the problem that the sentiment orientation calculation in traditional methods is affected by the position of seeds,a method of calculating the sentiment value based on the label propagation algorithm is proposed to obtain a higher-precision domain-specific sentiment lexicon.Experiments on two domain datasets show that the proposed method can effectively improve the F-measure of the domain-specific sentiment lexicon.(2)Research on word embedding model of domain-specific sentiment lexicon in fusion and text sentiment analysis method based on sentiment word vectors and long short-term memory network.For the problem that word-embedding model word2 vec has similar semantic but inconsistent sentimentality when measuring similar words,a word-embedding model combining sentiment information and semantic information is proposed by integrating domain-specific sentiment lexicon with sentiment information dimension,to improve the sentiment expression of word vectors.Aiming at the problem that traditional machine learning method ignores the semantic relationship of text,a text sentiment analysis method based on sentiment word vectors and long short-term memory network is proposed to make full use of text depth semantic information to improve the effect of sentiment analysis.Experiments on word-level and document-level sentiment tasks show that the proposed word embedding model has stronger sentiment expression ability,and the proposed sentiment analysis method can effectively increase the F-measure.(3)Research on the application of text sentiment analysis in satisfaction of the automotive customer.The paper studies the modular hierarchical structure by the fusion of text sentiment analysis and network technology,designs and develops data collection module,data preprocessing module and data analysis and statistic module to realize the automotive customer satisfaction system.Based on the results of textual sentiment analysis,the system visually displays the satisfaction of various models of car and car properties.The validity and practicality of the proposed model and method are validated in the application of automotive user satisfaction.