Managing information in online product review communities: A comparison of two approaches

Virtual communities often suffer from a number of problems, including questionable information quality and information overload, which threaten their utility and stability. To address this, social filtering techniques may be used, in which users rate the postings, guiding others to the important ones. This method is contrasted to information retrieval techniques, in which intrinsic properties of texts, such as length or keywords, are used to rank them by perceived relevance to a topic. Each approach has advantages and disadvantages. Social navigation assumes that users actively rate messages, however, soliciting sufficient participation is a known challenge. Additionally, what is interesting for one user may not be for others. Currently, we compare these approaches in the context of an e-commerce product review forum at Amazon.com. We find that while a significant proportion of reviews go unrated, these reviews are typically of low quality. Interestingly, we also find that the rankings produced using reader-assigned "helpful votes" are correlated to the rankings assigned by some simple information retrieval algorithms. The conclusion is that a number of approaches for filtering product reviews could effectively be used in such online communities in order to accommodate user preferences, and thus, in reinforcing the utility of the community.