Research On Temporal Characteristics Of Online User Review And Its Impact Based On Content Mining

With the emerging of Web2.0, The notion of user participation is growing in popularity. User generated content(UGC) gradually dominate the way of content generation on website. As one of the most important form of user generated content, online user review is a kind of product or service evaluation which was voluntarily posted on the company or third party platform by consumers. Online user reviews have been proved be very important to reviewers themselves, potential consumers, e-commerce platform and manufactures.This thesis involves three kinds of work on temporal characteristics, impact and content of online user reviews.(1) Exploring when will consumers post their online reviews after they finish purchasing?(2) Mining what kind of content did reviewers post in their reviews?(3) Exploring why did consumers post this kind of content in such a time interval?In order to explore when will consumers post their online reviews after they finish purchasing, we employed theories and methods from human behavior dynamics to depict temporal characteristics of consumer online buying and reviewing behaviors. With the experiment result on real dataset, the “purchasing- reviewing” time interval follows a power-law distribution. By dividing time interval sequence of “purchase and review” into stages, we found that consumer in different stage behaved differently.For the task of mining out what kind of content reviewers posted in their reviews, we tried to use text mining methods to extract information from massive online user reviews. Due to the inhabit feature of online reviews, such as massive, sparse and the poor semantic comprehensive result obtained by traditional text mining methods. We proposed a new content mining method of online reviews based on word embedding. The result of feature extraction and clustering from online reviews on real dataset revealed the effectiveness of the method and generated a good comprehensive result. The result of event detection in social media verified its generality of our proposed method.For the task of exploring why consumers post this kind of content in such a time interval, we tried to find out what factors will influence promptness of consumers’ review. Combined with consumer motivation theory and social exchange theory, this thesis proposed a theoretical concept framework from which the consumer’s own motivation, electronic commerce platform, social relations and consumers’ product experience will influence consumers’ review promptness. Then a “review-feature” mining and mapping method was proposed to extract the variables from massive online user reviews. Finally, the result of extracted features by text mining and the regression method on real review dataset validated proposed hypothesis. The results show that membership, operating experience of products have a positive effect on review promptness. Price, Rating polarity, service, logistics, appearance, operating system and cost-effective have a negative impact on review promptness.The contribution of this thesis in mainly reflected in the following three aspects.(1) The temporal characteristics of two typical behavior(purchasing and reviews) in electronic commerce was systematically depicted with methods of human behavior dynamics. The time interval between purchasing and reviewing follows a power law distribution, which provides new empirical evidence for the research of online human behavior dynamics.(2) This thesis proposed a new mining method based on word embedding and a new semi-supervised “review-feature” mapping method. With these two methods, electronic commerce companies are able to efficiently extract information from massive online reviews and solve the problem of poor accuracy and comprehensive by conventional mining method from massive sparse short review text.(3) This paper employed consumer motivation theory and social exchange theory to build a theoretical model of influence factors on review promptness from the perspective of consumers’ motivation, e-commerce platforms, product and consumers’ society tie. And Text mining method and econometric modeling method was combined to validate the proposed model.