We suppose that all we know about the preferences of an agent, is given by a (small) collection of relative preferences between choices represented by their evaluations on a set of criteria. Taking lesson from the success of the use of analogical proportions for predicting the class of a new item from a set of classified examples, we explore the possibility of using analogical proportions for completing a set of relative preferences. Such an approach is also motivated by a striking similarity between the formal structure of the axiomatic characterization of weighted averages and the logical definition of an analogical proportion. This paper discusses how to apply an analogical proportion-based approach to the learning of relative preferences, assuming that the preferences are representable by a weighted average, and how to validate experimental results. The approach is illustrated by examples.
CITATION STYLE
Pirlot, M., Prade, H., & Richard, G. (2016). Completing preferences by means of analogical proportions. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9880 LNAI, pp. 135–147). Springer Verlag. https://doi.org/10.1007/978-3-319-45656-0_12
Mendeley helps you to discover research relevant for your work.