Integrating Bayesian network classifiers to deal with the partial label ranking problem
Proceedings of The 11th International Conference on Probabilistic Graphical Models, PMLR 186:337-348, 2022.
The label ranking problem consists in learning preference models from training datasets labeled with a (possibly incomplete) ranking of the class labels, and the goal is to predict a ranking for a given unlabeled instance. In this work, we focus on the particular case where the training dataset and the prediction given as output allow tied class labels (i.e., there is no particular preference among them), known as the partial label ranking problem. This paper transforms the ranking with ties into discrete variables representing the preference relations (precedes, ties, and succeeds) among pairs of class labels. We then use Bayesian network classifiers to model the pairwise preferences. Finally, we input the posterior probabilities into the pair order matrix used to solve the corresponding rank aggregation problem at inference time. The experimental evaluation shows that our proposals are competitive in accuracy with the state-of-the-art mixture-based probabilistic graphical models while being much faster.