How communities behave

Here is a clip from a note that arrived in the ACM email newsletter Technews the other week:

A researcher who studied the voting patterns on Amazon.com, the Internet Movie Database (IMDb), and the book review site BookCrossings says a small group of users can easily affect online recommendations. In examining hundreds of thousands of items and millions of votes across the three sites, Carnegie Mellon University (CMU) professor Vassilis Kostakos discovered that a small number of people were responsible for most of the ratings. His team found that only five percent of active Amazon users voted on more than 10 products, and a handful made recommendations for hundreds of items. Kostakos believes such voting patterns may not reveal views that represent the overall community, and notes that the results help explain why reviews are often overly negative or overly positive. He believes Web sites could get more users to participate by making it easier to offer reviews.

In some way I guess it is great to finally get hard scientific facts behind what I always believed anyway. I tend to use 1-10-100 as a rule of thumb when I think about online communities: 1% of the members take significant responsibility for the community, 10% contribute significant content, and 89% mostly read and watch.

Note, though, that the role of the 89% is no less important for the social dynamics of the community than the apparently more active roles. And this may be where Professor Kostakos in the quote above goes wrong. Believing that voting patterns should reflect the views of the overall community, and thinking that more people would participate actively if it was easier to offer reviews, seems misguided to me.

My perspective is rather that online reviews and recommender system output are not supposed to be representations of a majority view, and that community members generally know this.