A Study On The Impact Of Online Reviews On Product Sales

The continuous progress of Internet technology has fully driven the rapid development of e-commerce,and more and more people tend to spend much time on online shopping.They always express their experience and feelings about the good or service which their buy,which makes massive amounts of information emerged on the various of e-commerce websites.Not only does online review become a new network communication means and platform,but also it has a great influence on sales.However,most researchers who do research on the analysis of the influence factors of product sales only focus on some variables such as the volume of online reviews and the readability of comments,without taking the attributes of products and the sellers'responses into account.Based on this reason,on the basis of data about the game notebook from the Tmall,this paper first studied the attributes of product and its related sentiment intensity,and verified the types of business feedback with the help of topic models subsequently.Then,this research built mathematical models to analyze the influence of this two kinds of factors on the product sales.The main contents and conclusions about this paper can be drawn as follows:(1)This paper proposed a new kind of fine-grained sentiment analysis method based on syntactic analysis and deep learning,which is used to extract product attribute words and its related sentiment words.In terms of feature extraction,this method combines the feature extraction algorithms based on high frequency nouns and semantic relations,and calculates the semantic similarity between product attribute words with the help of word2vec.Then,each attribute word should be classified into the corresponding product feature category in the method of cluster analysis.With respect to aspect sentiment calculating,on ths basis of sentiment lexicon and word vector model trained,this method can get some so-called implicit sentiment words which express sentiment orientation but don't exist in the sentiment lexicon,and calculate the sentiment intensity of each product attribute according to the semantic relations and product attribute words.From the point of view of methodology,not only did this method not only take the semantic dependency relations into consideration for making the selection of product attribute words more comprehensive and accurate,but also it calculates the semantic similarity in the way of word vectors.(2)This paper verified the kinds of business feedback in the method of probabilistic topic models.In this study,LDA algorithm was used to extract the meaningful topics from the text data about business feedback,and these selected topics were classified into relevant categories by analyzing the topic words.It could be founded that most sellers response to consumers with appreciation,apology and explanation,while the category of compensation is almost non-existent.By the verification of several common types of management response by means of machine learning algorithms,this study analyzed the sellers' responses in a more detailed way,providing a new perspective for related research.(3)This paper studied the influence of the attributes of products and the types of business feedback on the product sales.Based on the mathematical model whose independent variables are the price,the volume of online reviews and the number of sellers' responses,and dependent variable is the product sales,this study introduced the variables which stand for the features of product and the kinds of management responses as independent variables respectively,for exploring the role of these factors.The results showed that,first,the volume of online reviews has a positive effect on product sales,the effect of the number of sellers' responses is negative,while the effect of price is not significant.Second,for the influence of management response types on product sales,the impact of all three kinds is significantly positive,and responsing with apology has greatest effect where the affect of responsing with appreciation is least.Besides,for the influence of product attributes on product sales,not all the attributes will significantly affect the sales.Taking the game notebook as example,the features of heat radiation/sound,battery life,hardware and screen don't have significant effect,where others can affect the sales of game notebooks positively.