Research On Sentiment Text Classification For Product Reviews

With the development of e-commerce,the information of product reviews on the website is increasing day by day.For the purchased products or services,consumers can express opinions which can reflect the quality of them from different aspects According to the online product reviews,consumers who want to buy the products can know the required information about those products,which merchants can improve the weaknesses of the products and services in return.As the information commented by consumers is messy,it is necessary to analyze the sentiment text in order to facilitate other consumers for better understanding the product information and to help the sellers get the feedback in time.The analysis of text sentiment mainly focuses on analyzing the emotional features of the texts.In order to effectively extract the sentiment features of text,this thesis use feature selection algorithm and emotional dictionary,and then classify the text.The main contents are as follows:(1)The method of n-gram feature extraction and redundancy reduction based on chi-square statistic.Aiming at the problem that n-gram characteristic items are redundant and affect the actual classification effect,the traditional chi-square statistic is improved.The method not only utilizes the association between n-gram features to select features,but also utilizes the relevance between the features and different classes to determine whether the features are redundant or not,so that the higher correlation and lower redundant n-gram features can be selected.By using Support Vector Machine classifier to test the text orientation in different sentiment corpus,the experimental results show that this method is effective in text sentiment classification.(2)The method based on the emotional dictionary,emotional features can be directly extracted from the text.However,the quality of emotional dictionary will affect the result of classification,and the contextual structure of the modified emotional word will also affect the polarity of the emotional word in the text.Aiming at the problem that emotional dictionary construction is difficult and the polarity change of emotional words,thesis presents a method to sentiment classification based on product attributes.In this method,we use the word2vector training features to generate word vectors,and the similarity between word vectors is also used to cluster the similarity features.Using the dependency relation between attribute words and emotional words,extract attributes words and emotional words.Then,we not only analyze the features of sentiment text and construct the emotional dictionary,but also extract the attribute words,the emotional words and their contextual structure.Finally,classifying the text by using SVM algorithm and analyzing the impact of the method on the text classification,we can draw the conclusion that this method is effective.Based on that,this thesis analyzes the influence of LDA topic features on text sentiment classification.Considering the structural information of emotional features,the n-gram model is used to generate the n-gram features,and the redundant features are reduced.Then,the LDA topic probability is used as the feature,and the SVM algorithm is used to test the different sentiment corpus,and then the influence of different.n-gram features and LDA topic model on the text classification is analyzed.Finally,comparing the method with the different classification methods,it shows that this method improves the result of text sentiment classification and verifies the effectiveness of the method.