[noob]: Creating preferences list with "not always correct" features

I have the following problem: - A set of 10 classes (for example: dogs, cats, rabbits,...) - A set of 1000 items (pictures of these animals) - For every item I have its class (is it a dog picture, or a cat picture, or a rabbit picture) - There is a 30% chance that the associated class for an item is wrong (for example: a dog picture we think is cat) - All items have counter how many people clicked on them

I show to a user 7 pictures, and after he clicks on some, i should update his profile in order to show him better pictures next time. My current approach is to define a float per category for every user, and use that float in ranking 1000 items. Choose top 7, and when he selects one, increase the number for that category, and decrease for others. Any ideas on how to work with this? Thanks!