| Online product reviews on the shopping site are an important manifestation of UGC(User Generated Content),which is a great help for assisting consumers in purchasing decisions and helping businesses improve their goods and services.However,on the other hand,with the Internet technology rapidly penetrating people's life,online reviews and other UGC also showed an explosive growth.The quality of these online reviews is uneven,making it difficult for consumers to find the information they want from a large number of comments.In order to meet consumers'special information needs for the specific product details and to dig out high quality online reviews,this paper try to rank the online reviews of different subjects by taking an example of the travel reviews of Meituan.com.By extracting the topics of the online reviews,we can help the consumer to quickly locate the desired information;through a sort of sorting method,we can filter out high-quality online reviews.In order to explore what kind of online reviews are high quality and useful for consumers,this paper has conducted a study of the factors that influencing the online reviews usefulness.Based on around the 720,000 online travel reviews of Meituan.com,and taking the helpful votes of online reviews as dependent variables and take the history review information,information quality of online reviews and extremity of online reviews as independent variables,we build the online reviews factor model.In this paper,we use zero expansion negative binomial model instead of the traditional linear regression model to analyze the data.The results show that the length and the picture numbers of the online reviews,the number the reviewers had published and the useful votes the reviewer had received,the polarity of the online text reviews have positive influence on online review's vote number.The score of the online review has no influence on increasing the amount of the review's vote,but help to explain why the votes of the online reviews is 0.Compared with the low score reviews,the high score reviews have a bigger chance to get a zero vote.Finally,based on the results of data analysis,we determine the online reviews usefulness evaluation model.In order to extract the topics of the review,this paper carries out the topic extracting study of online reviews text based on the LDA model.Besides the word segmentation and modeling,the best topics choosing problem and parameter setting of LDA,we also discuss other import questions that rarely mentioned in the other theme extracting research papers.To deal with the spam reviews,we give two solutions to identify them.Aiming at the problem that the quality of the topics trained by LDA is uneven,we use two methods named UCI and UMass to screen the high quality topics from the view of topic coherence.The results show that both UCI and UMass methods have a good effect on screening topics with high quality.Finally,we take the reviews of the Wuhan Happy Valley as the experimental object to test our topics extracting and online review sorting method.To validity the usefulness of our method,we use questionnaire to invite consumers to score the online reviews and compare it with the score given by our method.The results of the questionnaire show that our online review ranking method based on the relevance of the topic and the usefulness of the comment can help to select the useful reviews on under each topic. |
PDF Full Text Request