Research On The Influencing Factors Analysis And Prediction Of Extreme Review Helpfulness Of Search Products On E-commerce Platform

With the rapid growth of e-commerce websites,the number of online forums is increasing.Online reviews have been playing a critical role as consumers choose to share their perceptions and opinions of products or services.According to the review polarity,online reviews can be divided into either extreme reviews(positive and negative reviews)or neutral reviews.Recent evidence suggests that consumers think that extreme reviews may be more useful than neutral reviews.Not all extreme reviews are helpful to consumers yet,and tedious extreme reviews are causing information overload.If the managers of e-commerce platform could intelligently recommend high-quality reviews to consumers,it will not only help consumers make purchasing decisions,but also make profits for e-commerce platforms.Efficiently identifying useful reviews has become a critical research issue.Therefore,based on the bibliometric analysis of the literature of online reviews at home and abroad,the research data sets were extreme reviews of current mainstream mobile phones on Jingdong Mall,consisting of 14561 positive reviews and 9721 negative reviews.The research object was extreme review helpfulness,we employed text mining and natural language processing(NLP)to deeply explore the quantifiable features of online reviews.We constructed the model of influencing factors of extreme reviews helpfulness,and adopt tobit regression analysis to test hypotheses.Besides,eight machine learning algorithms were used to predict the helpfulness of extreme reviews.The results indicate that,posting photos are the determinant factor for positive reviews.Highly helpful positive reviews are those reviews with a long review length,a long effective review length,photos posted and a low emotional intensity.Posting videos is the most critical factor for negative reviews.Highly helpful negative reviews are those reviews with a long review length,a long effective review length,photos posted,videos posted,and low emotional intensity.For the helpfulness prediction of positive reviews,Naive Bayes has the best prediction effect,followed by the neural network;For the helpfulness prediction of negative reviews,the neural network has the best prediction effect,followed by naive Bayes.This study broadened the theoretical depth of the research of exploring influencing factors of online review helpfulness,and enriched the research variables of review helpfulness,to make the research of review helpfulness in a dynamic way.Besides,the perspective of helpfulness prediction was different from previous studies.This research helps e-commerce platforms to better understand consumers' online comment behaviors,to implement effective online comment management strategies,enhance the electronic word-of-mouth effect and create a good consumer shopping experience.