An Lda and Synonym Lexicon Based Approach to Product Feature Extraction from Online Consumer Product Reviews
Baizhang Ma, Dongsong Zhang, Zhijun Yan, Taeha Kim · Journal of electronic commerce research · 2013
ABSTRACTConsumers are increasingly relying on other consumers' online reviews of features and quality of products while making their purchase decisions. However, the rapid growth of online consumer product reviews makes browsing a large number of reviews and identifying information of interest time consuming and cognitively demanding. Although there has been extensive research on text review mining to address this information overload problem in the past decade, the majority of existing research mainly focuses on the quality of reviews and the impact of reviews on sales and marketing. Relatively little emphasis has been placed on mining reviews to meet personal needs of individual consumers. As an essential first step toward achieving this goal, this study proposes a product feature-oriented approach to the analysis of online consumer product reviews in order to support feature-based inquiries and summaries of consumer reviews. The proposed method combines LDA (Latent Dirichlet Allocation) and a synonym lexicon to extract product features from online consumer product reviews. Our empirical evaluation using consumer reviews of four products shows higher effectiveness of the proposed method for feature extraction in comparison to association rule mining.Keywords: Online product reviews; Feature extraction; Latent Dirichlet Allocation; Synonym lexicon; Data mining(ProQuest: ... denotes formulae omitted.)1. IntroductionWith the rapid advance of Internet technology, online shopping becomes increasingly popular. iResearch reports that the e-commerce market worldwide reached US$1.99 billion in Fall 2012, which was 21.9% more than 2011 [iResearch 2012], However, online shopping makes it impossible for consumers to obtain the first-hand experience and knowledge about product quality through direct touching, seeing, and trying out products before purchase.With the rise of Web 2.0 technology, users have become more and more comfortable with sharing their opinions on products or services over the Internet. Examples of these Web 2.0 web sites include e-commerce sites such as Amazon.com, social network sites like Facebook.com, local service provider review sites like Yelp.com, and digital media distributors like Netflix.com. Empirical studies have shown that user reviews indeed have significant impact on sales, and users rely much more on reviews than on numeric ratings to decide if a product should be purchased [Chevalier & Mayzlin, 2006, Li et al., 2009], Consumers often rely on other consumers' online reviews of features and quality of products while making purchase decisions because those reviews offer more useful information about products than the descriptive information available on a shopping website provided by manufacturers [Park and Kim 2008, Hu et al. 2009], According to a survey, 70 percent of Americans regularly consult online consumer product reviews or ratings before making an important purchase [Ante 2009],The rapid growth of the volume of online consumer product reviews, however, makes browsing a large number of reviews to identify information of interest time consuming and cognitively demanding. It poses a significant challenge for taking advantage of those reviews. Review mining that uses data mining and text mining techniques for analyzing a large quantity of online consumer product reviews [Popescu and Etzioni 2007] has been extensively explored as a means of addressing this information overload problem. The majority of existing research, however, focuses on prediction of helpfulness of reviews, determining the polarity of consumer opinions on products, and the impact of reviews on sales and marketing, etc. Relatively little emphasis has been placed on mining reviews to understand consumers' opinions toward individual features of specific products. Product features include product attributes, components, component attributes, etc. [Xi et al. 2011],In reality, individual consumers may have different preferences for features of a product. …