Are Online Product Reviews Trustworthy? Automatically Detecting Bilingual Deceptive Product Reviews and Analyzing Their Economic Impact

Description

Have you ever thought that some user-generated hotel reviews at ctrip.com or product reviews at amazon.com are fake? In the 2009 Nielsen Global Online Consumer survey, which involved 25,000 respondents from 50 countries (including China), 70% of the respondents said that they read online reviews before making purchases. 1 Nevertheless, there have been growing concerns about the trustworthiness of online product reviews. Firms have financial motivations to generate deceptive reviews (i.e., fictitious reviews that are written to deceive the readers) that inflate the true qualities of their products, or defame their competitors’ products. Earlier this year, the Federal Trade Commission in the U.S. fined a company selling a series of guitar-lesson DVDs $250K for deceptively advertising its products through online affiliate marketers who falsely posed as ordinary consumers.2 The Daily Background also reported that a computer retailer was paying 65 cents per review to people who posted “shill” reviews promoting the company’s products at amazon.com.3 Deceptive online product reviews, which are supposedly boost the perpetrator’s sales, can bias consumers’ purchase decisions, and may have a detrimental effect on other vendors’ sales. To promote fair trading and maintain appropriate consumer welfare in the Greater China area, there is a pressing need to develop effective and efficient deceptive review detection services to identify and filter out bilingual deceptive reviews on the Web.It is not feasible to rely on a manual approach to detect deceptive reviews for two reasons:Humans are poor at deception detection because of the “truth bias” andhuge volumes of product reviews are posted to the Web every day.However, automated detection of deceptive reviews is a very challenging research problem. Unlike the detection of Web or email spam, where discriminatory features (e.g., some commercial spam keywords) are available to a detector, explicit features are often missing in deceptive reviews. Deceivers deliberately hide the traces and make a deceptive review look just like a legitimate one. Existing methods for detecting deceptive reviews often employ a deviation-based (e.g., detecting the rating of a review deviated from the average rating) or similarity-based (e.g., detecting the high content similarity between two reviews) approach. Nevertheless, the former approach suffers from its poor reliability, as the content or rating of a review may deviate from that of the majority due to genuine individual preferences. The latter approach suffers from a high degree of computational complexity during detection time, and is therefore difficult to scale up to process the huge volume of online product reviews. Furthermore, most existing methods focus on the detection of deceptive product reviews written in English only.Guided by the Design Science research methodology and grounded in Interpersonal Deception Theory, the aims of the proposed research project are as follows.To address the shortcomings of existing methods for detecting deceptive reviews, we will design a bilingual deceptive review detection service that can efficiently “mine” the context-dependent, latent traces left by deceivers and thus improve detection accuracy across multiple deception contexts.To provide consumers with proper advice related to review-based purchase decision making, we will use our deceptive review detection service to conduct the first large-scale empirical study of the trustworthiness of online product reviews in bilingual contexts. The data set will include tens of millions of reviews crawled from major Web sites such as taobao.com, amazon.com, cnet.com, etc.To assess the detrimental effect of deceptive reviews on firms’ sales and consumers’ purchases, we will conduct the first econometric analysis of bilingual, deceptive reviews using the results of the aforementioned deceptive review detection process and the actual sales data of some products.The deceptive review detection service designed by this project can be adopted by e-Commerce Web sites or third-party agencies that monitor reviews (e.g., Reputation.com) to improve the overall quality and hygiene of online product reviews in bilingual contexts. The proposed research project will foster fair trading and enhance consumer welfare in the Greater China area.