Completing preferences by means of analogical proportions

Publisher:
Springer
Publication Type:
Conference Proceeding
Citation:
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) Modeling Decisions for Artificial Intelligence 13th International Conference, MDAI 2016 Sant Julià de Lòria, Andorra, September 19–21, 2016 Proceedings, 2016, 9880 LNAI pp. 135 - 147
Issue Date:
2016-01-01
Full metadata record
Files in This Item:
Filename Description Size
21062017111601-0001.pdfPublished version614.77 kB
Adobe PDF
© Springer International Publishing Switzerland 2016.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.
Please use this identifier to cite or link to this item: