Research On Online Product Reviews Emotion Analysis And Comment Usefulness

A large amount of comment data appears on various websites,which hides a lot of valuable information in them.We can employ fine-grained emotional analysis of comment data to explore users' emotional tendencies towards products,discover product strengths and weaknesses,and provide strong data support for improving product quality.We can also make useful of results of analyzing on these comment data,so that customers can find useful comments in a large number of comments,then quickly complete the completion of shopping decisions,and save time for customers.So how to analyze the user's emotion from the comment data and find the useful comment has become urgent.In order to solve these problems,this paper adopts two different ways in the research of commodity comment emotion analysis,one is based on emotion dictionary,the other is based on machine learning.In the research on the usefulness of comments,the research is based on the usefulness of commodity attributes.The main work of the study is as follows:1.We researched related literature review based on dictionary emotion analysis,machine learning emotion analysis and research of usefulness of comments.Compared with the advantages and disadvantages of previous studies.2.In the process of emotion analysis based on dictionaries,a relatively complete basic emotion dictionary is constructed.The dictionary of emotional intensity adverbs with different affective intensities was established,and the emotional intensity adverbs are divided into fivegrades.A dictionary of network terms and a dictionary of negative words are also constructed.Finally,the modified SO_PMI model was used to calculate the emotional tendency of comments,analyzed the comment from the emotional value,the average value of emotion,and the difference of emotion standard,and then the dictionary-based emotional analysis was completed.3.In the process of emotional analysis based on machine learning,1000 items of positive and negative affective tendency data are first marked respectively,and then the three common machine learning methods was conducted experimental comparison including monadic grammar and binary grammar and monadic grammar + binary grammar(Bayes,support vector machine,logical regression),and the accuracy of different Bayesian decisions for classification was analyzed.The accuracy of different SVM kernel functions for classification was analyzed.The optimal feature engineering was determined,selected the appropriate word vector dimension,and the optimal classifier was determined.Finally,the accuracy of the polynomial Bias classifier was up to 93.0%.4.In the analysis of the usefulness of commodity comment,we first use the word2 vec neural network model to build a commodity attribute dictionary.The attribute dictionary contains 11 attributes of the mobile phone.Under each attribute,there are 100 description words about the properties of the attribute with a total of 1,100.Build a useful comment model(PA_SO_PMI),analyze the usefulness of the comment by commenting on the usefulness value,the average value of the usefulness,mean square devia-tion,and get the most useful 20 comments by sorting the mean square devia-tion of the usefulness,thus making users easy browsing.