Skip to Main Content
Link structures among items within an E-commerce Web site can be regarded as a potential recommendation that helps new consumers quickly locate relevant products. In this paper, combining a modified version of Google's PageRank method with economic analysis of word of mouth, we investigate whether a product's position within a network composed of recommendation links is incrementally informative about its future sales. Based on data from Amazon.com, we document that with consumer word of mouth and other product characteristics controlled, the position of a product within a recommendation network does influence consumers' purchase decisions, and models incorporating link structure have a higher incremental predictive power of future sales than models without. In addition, as time elapses, the relative weights consumers placed on recommendations with price discount and those without are different. Last, we develop a learning mechanism through which we find the optimal damping value of the PageRank IR model in the Amazon context. Our results show that compared to general Internet surfing behavior, consumer consumption on Amazon is less random. We conclude that even though the product position within a recommendation network does influence customers' purchase behaviors, product sales are still mainly driven by their own product characteristics.