I totally agree with Joshua Porter about Amazon’s new review interface - it’s awesome. It provides even greater insight into user ratings by showing a breakdown of the number of people who selected each star rating. Not only is it awesome in and of itself, it also seems easily extensible to allow for a more personalized view of user ratings. Since Amazon knows so much about its users, both the person seeking ratings information and the people who have provided ratings, I could see another option that shows how people like you rated the current item.
In the current interface, a user can rollover the star rating and see a bar chart showing the breakdown of ratings for all users, but it might also be interesting to see how the breakdown changes when only viewing the ratings of users like you:

For particularly polarizing items - ones that users seemed to either have loved or hated - this kind of insight could be very helpful. Netflix does something similar to this for their movie ratings, by showing you overall ratings as well as “Our best guess for you.” I find this pretty helpful and often quite accurate, but I also have to admit that I think the web is floating in a sea of increasingly meaningless 5-star ratings and it sucks to have to add to the clutter.
I thought for awhile that we should get rid of 5-star ratings altogether - giving users the option to say they love or they hate it should be good enough. However I was surprised to read over in the Netflix Community Blog that users were demanding even more granularity for their ratings - they really want to be able to rate with half stars. That’s when it occurred to me that if you’re passionate about something, you probably want to be as specific as possible about how you feel about it. Those same users are probably fine with StumbleUpon’s thumbs up or down model, but when it comes to movies…they really want to express themselves. So although this level of specificity might pass the casual user by, I think the passionate ones (who are often also the ones creating the content!) will appreciate it.
September 21st, 2007 at 9:28 am
Great post and i agree on more information making more sense. An affinity meter would also be great (we are about to launch one for each club on u-lik)
If you take the netflix prize sample, you have the following split in ratings :
1 = 4,6% / 2= 10.1% / 3 = 28.7% / 4= 33.6% / 5 = 23.1 %
And since the RMSE for moviematch is around 1 it clearly means that they have no information at all.
But, the worst is probably that going more granular is not helping even if it can make your power users happy.
October 4th, 2007 at 2:59 pm
Yeah, I like the new Amazon break-out, too. It looks like they’ve taken a cue from Summize (that uses Amazon’s own data, too.) Check out the wealth of specificity here:
http://www.summize.com/product/mulholland-drive/dvd