Using Properties of the Amazon Graph to Better Understand Reviews

According to a 2010 paper published by PowerReviews, a e-tailing consulting company, 63% of shoppers consistently read reviews prior to making a purchase decision while 33% of that total spend at least half an hour reading reviews for a product; thus online reviews of a product have now become a strong factor in determining the sales of a particular product. Perhaps no retailer is more influenced by online ratings than Amazon.com, the world‘s largest online retailer, and one of the pioneering companies of online feedback scores. Amazon.com also has the world‘s largest collection of online reviews [1], which further emphasizes the importance of product reviews for the online retailer. However, there has been significant controversy surrounding the accuracy and origins of many of these reviews. A 2011 article by Daily Mail, for instance, reports that that many manufacturers have started hiring companies to post fake reviews on Amazon.com in an attempt to increase the manufacturer‘s credibility while at the same time, defaming competition [6]. Trevor Pinch, a professor of sociology at Cornell, along with Web entrepreneur Filip Kesler published a paper further strengthening the assumption that there is personal bias involved in many of the reviews, by uncovering the fact that around 85% of Amazon‘s top 1000 reviewers receive free products from publishers, agents, authors, and manufacturers [2]. The helpfulness rating of the reviews themselves have also been found to be bias. [3] demonstrates that user helpfulness ratings oftentimes do not directly reflect the content quality of the review itself. [4] elaborates on this notion by illustrating the fact that users rate the helpfulness of a review based upon their own personal preference towards the item. This project aims to accomplish two tasks: to be able to predict future Amazon ratings given current rating data, and to adjust the current star rating scheme to better reflect the true quality of the item. Predicting the future rating of a product is valuable because it gives both manufacturers and customers the opportunity to know the consensus perception of the product at an earlier time. Being able to establish when the future rating can be predicted with a high degree of confidence is also desired since many customers will not purchase a product until they are confident that the consensus view of the product is a favorable one. In fact, [1] found that 72% of consumers find user product reviews to be a “very/extremely” important factor when it comes to selecting and purchasing a product. Because a vast majority of users rely upon Amazon.com reviews to determine the quality of a product, attempting to produce more accurate ratings by accounting for the biases mentioned above is also an important task. The star rating likely represents the first opportunity most users have to pass judgement on an item they find when browsing Amazon, and the star rating also factors into other components which can influence the sale of a product, such as when that product appears in the search results.