A Method of Preference and Utility Elicitation By Pairwise Comparisons and its Application to Intelligent Transportation Recommendation Systems
Alexander Borodinov, Anton Agafonov, Vladislav Myasnikov · 2020
The paper is devoted to the object ranking problem, one of the preference elicitation tasks considered in machine learning. We analyze this problem in the formulation when it is necessary to reconstruct the utility function using observations. Each observation is the pairwise comparison result of utility function values for two random variables. We propose a unified approach to the utility function elicitation problem. This approach consists of two steps: adaptive transformation to the high-dimensional space and construction of linear (for the utility function elicitation) or non-linear (including non-parametric, for the preference elicitation problem) classifiers. In experimental analysis, we compare different methods of transformation to the high-dimensional space and different classification algorithms. We investigate their effectiveness to solve model and real problems of the utility and/or preference function elicitation problem.