Study On Sentiment Orientation Analysis Of Product Reviews Based On Presentation Skills

With the rapid development of internet technology and the increasing popularity of electronic commerce,online shopping has gradually become a normal life in daily life.Large amount of online reviews have emerged during online shopping process,and the sentiment analysis of these reviews can not only provide useful information for consumers,but also help enterprises optimize business decisions.At present,some research results have been achieved in the sentiment analysis of product reviews.However,the existing methods are of low classification precision and small application scope.The two main methods of sentiment analysis are artificial semantic rule based methods and machine learning methods based on corpus.This thesis makes a further study on the artificial semantic rule based methods and proposes a new sentiment analysis method based on presentation skills to improve sentiment classification accuracy without destroying the cross-domain characteristic.Then the two main methods of sentiment analysis are combined to further improve the sentiment classification accuracy in the determined field.The main research work consists of three parts:(1)Based on the coding and decoding theory of information transmission,this thesis proposes a sentiment analysis method based on presentation skills.This method separately designs the schemes of recognizing presentation skills and grading emotional score of presentation skills according to the different skills used by comment creators to express the emotional tendencies.In the field of recognizing presentation skills,we use two kinds of keywords matching methods to recognize presentation skills.One is named as simple keywords matching and the other one is called compound keywords matching.The choice of the matching methods is decided by the structural characteristics of presentation skills.In the field of emotional calculation,the emotional calculation methods are divided into direct calculation and indirect calculation according to the recognition accuracy of presentation skills.Finally,the emotional tendency of the review is judged by the average emotional score of each clause calculated by its presentation skills.(2)This thesis perfected and improved the traditional sentiment analysis method based on emotion lexicon.First,on the basis of the three main emotional etymology of HowNet,NTUSD and the emotional vocabulary noumenon of Dalian University of Technology,the emotion lexicon is expanded by adding "dictionaries ofcommendatory words" and "dictionaries of derogatory words" as etymology.Next,in the process of emotional calculation,it is taken into extra consideration of the influence of negative verbs on sentiment orientation and the different collocations of negative words and degree adverbs.(3)In view of the deficiencies of the sentiment analysis method based on presentation skills,we propose two improved strategies that combines presentation skills with SVM,PSAS and PSBS.The PSAS combines presentation skills and SVM with weight sums,and the weights come from the classification accuracy of the two methods.The PSBS takes presentation skills as the features of feature extraction,and then uses SVM training classifier to implement sentiment classification.The experimental results show that the two combination methods can both improve the classification performance of the sentiment analysis method based on presentation skills.