Research On Sentiment Analysis Methods For Tourism E-Commerce Text Comments

Sentiment analysis classifies text according to the emotional information expressed in the comments.However,e-commerce text comments contain users’ subjective evaluation of the services and attributes of the goods.In some cases,the overall emotional tendency of the text cannot directly obtain the emotional tendency of a certain attribute of the goods.Therefore,this paper takes the text reviews of tourism e-commerce as the research object and conducts text-level sentiment analysis and aspect-level sentiment analysis respectively to meet the needs under different circumstances.The research content mainly includes the following aspects:(1)For the recognition of opinion information units,opinion information units are words used to express opinions and have emotional tendencies in text reviews,including evaluation objects,degree words,opinion words and emotional words.This paper selects a rule-based method for opinion information unit recognition.This method only considers the part-of-speech relationship,dependency syntactic relationship and semantic dependency between words in the sentence,And fully consider the part of speech of the verb,and adopt the circular recognition method to solve the complicated situation that the evaluation word is a verb phrase.(2)For text-level sentiment analysis,this paper proposes a combined with the characteristics of the dual channel convolution electricity text sentiment analysis model of neural network,to solve ignore context meaning between words caused by single convolution neural network,it is difficult to capture a long emotional information,and traditional word vector can’t identify the same words in different sentences for different components.Based on the result of viewpoint information recognition,this method introduces part of speech features and viewpoint information features as extended features.Then word vector and extended eigenvector are used as two inputs of convolutional neural network for sentiment analysis.Compared with other sentiment classification methods,the performance of the proposed method is improved greatly.(3)For aspect-level sentiment analysis,This paper divides it into two parts:the recognition of the evaluation unit and the quantification of the emotional tendency of the evaluation unit.In the recognition of evaluation unit,sentences are divided into key sentences and non-key sentences based on the recognition results of opinion information,and the influence of redundant text on the recognition of evaluation unit is reduced by filtering the hypothetical statements and declarative statements of key sentences.In the process of declarative sentence filtering,it is divided into concise and complex cases based on the opinion expression habit,and an evaluation object influence factor is introduced in the complex case.According to the results,this method has a good effect on the identification of evaluation units.In addition,the method of constructing evaluation collocation table combined with the structure diagram of comment objects is used to identify implicit evaluation units.Finally,we quantified the affective tendency of the evaluation unit based on the affective dictionary,and constructed a reverse dictionary to solve the context dependence of the evaluation words.Finally,the conclusion is given and the direction of the next work is prospected.