Determine the effectiveness of single-target multi-output regression for label ranking

Determine the effectiveness of the single-target multi-output regression approach, which independently predicts each rank position and sorts the resulting outputs to produce a total label order, for label-ranking problems, particularly whether it adequately captures dependencies between ranking positions.

Background

The paper situates its multi-output-regression framework in relation to a simpler single-target strategy previously used for label ranking. In that strategy, separate regressors predict the rank positions independently, after which the predicted values are sorted to obtain a total order. Unlike models that explicitly represent interactions among rank positions, this strategy disregards dependencies between targets.

The authors characterize the empirical effectiveness of this approach as unresolved because its independent treatment of rank positions may fail to capture dependencies among them. The question concerns the practical suitability of the single-target strategy for label-ranking prediction, rather than the central contribution of the paper, which develops and evaluates a broader framework for partial label ranking.