Handling Inconsistency in (Numerical) Preferences Using Possibility Theory

Gathering the preferences of a user in order to make correct recommendations becomes a difficult task in case of uncertain answers. Using possibility theory as a means of modelling and detecting this uncertainty, we propose methods based on information fusion to make inferences despite observed inconsistencies due to user errors. While the principles of our approach are general, we illustrate its potential benefits on synthetic experiments using weighted averages as preference models.