Rating by Ranking [more ideas]
Almost everything nowadays can be rated on various websites: books, movies, restaurants, videos, blog posts.. Sometimes we give a numeric value 1-to-10 or 0-to-100, sometimes we give stars out of five, other times just thumbs up or thumbs down.
The collected ratings can be used for various purposes: creating ranked lists (top 100 movies of all time, best restaurants in town, etc.), filtering out spam, or non-relevant items, or creating recommendation systems: "people who have rated items similarly to you have also rated highly the following ..." or "items that have been rated similarly by others to your favorite items are the following ..." or more complex algorithms.
In all these systems there is the problem that the information collected through ratings is subjective, noisy and people are usually not consistent in their rating strategies. For example someone might rate every movie with three stars, except those he doesn't like, which he rates with one star. Another user doesn't rate anything, except those he likes for which he gives five stars, etc. There are various strategies for dealing with these biases but the problem remains that the rating data is sparse and unreliable.
Part of a solution would be to collect more fine-grained feedback. However, this needs to be done in a way that doesn't need too much effort from the user. A possible approach would be to ask the user to rank items instead of giving numerical ratings. For example the user looks up "The Godfather" on a movie website and notes that he has seen the movie. Then he goes to the "Goodfellas", and there instead of rating, he is asked to indicate wether it was better or worse than the Godfather. When a new item is to be rated, there would be a drop-down list with the previously seen elements ranked by preference and the user would just need to indicate where the new element goes in the list by pointing at the correct location.
In practice, when the list grows too large, a fisheye-view list could appear, such as the one in this demo. Alternatively, the system could pick a subset of the total elements (say, 10 representative movies), and the user would just indicate where the new one fits in this smaller list. This would give less information, but it might still be enough for practical purposes.
With such a feedback system new algorithms are needed both for creating aggregate lists and for recommendation systems. Furthermore there has to be some possibility for the user to rearrange the ranking, to correct mistakes, etc. These personal rankings could add a new social aspect to many sites, where users could share their own ranked lists, compare/discuss each other's lists, etc.