A0423
Title: Multi-attribute preferences: A transfer learning approach
Authors: Sjoerd Hermes - Wageningen University (Netherlands) [presenting]
Joost van Heerwaarden - Wageningen University (Netherlands)
Pariya Behrouzi - Wageningen University and Research (Netherlands)
Abstract: A novel statistical learning methodology is introduced based on the Bradley-Terry method for pairwise comparisons, where the novelty arises from the method's capacity to estimate the worth of objects for a primary attribute by incorporating data of secondary attributes. These attributes are properties on which individuals evaluate objects in a pairwise fashion. By assuming that the main interest of practitioners lies in the primary attribute and that the secondary attributes only serve to improve the estimation of the parameters underlying the primary attribute, the well-known transfer learning framework is utilised. To wit, the proposed method first estimates a biased worth vector using data pertaining to both the primary attribute and the set of informative secondary attributes, which is followed by a debiasing step based on a penalized likelihood of the primary attribute. When the set of informative secondary attributes is unknown, we allow for their estimation by a data-driven algorithm. Theoretically, it is shown that, under mild conditions, the $\ell_\infty$ and $\ell_2$ rates are improved compared to fitting a Bradley-Terry model on just the data pertaining to the primary attribute. The favorable (comparative) performance under more general settings is shown by means of a simulation study.
