PROJECT TITLE :

Transductive Ordinal Regression

ABSTRACT:

Ordinal regression is often formulated as a multiclass problem with ordinal constraints. The challenge of coming up with correct classifiers for ordinal regression typically increases with the number of categories involved, due to the big number of labeled patterns that are needed. The availability of ordinal class labels, however, is often pricey to calibrate or troublesome to obtain. Unlabeled patterns, on the other hand, often exist in much bigger abundance and are freely offered. To take advantages from the abundance of unlabeled patterns, we gift a novel transductive learning paradigm for ordinal regression during this paper, specifically transductive ordinal regression (TOR). The key challenge of this paper lies in the precise estimation of both the ordinal class label of the unlabeled information and the choice functions of the ordinal classes, simultaneously. The core parts of the proposed TOR embody an objective operate that caters to several commonly used loss functions casted in transductive settings, for general ordinal regression. A label swapping theme that facilitates a strictly monotonic decrease in the objective function value is also introduced. Intensive numerical studies on commonly used benchmark datasets including the real-world sentiment prediction downside are then presented to showcase the characteristics and efficacies of the proposed TOR. Any, comparisons to recent state-of-the-art ordinal regression strategies demonstrate the introduced transductive learning paradigm for ordinal regression led to the sturdy and improved performance.