Efficient Regularized Least-Squares Algorithms for Conditional Ranking on Relational Data

TL;DR

This paper introduces a kernel-based framework for efficient conditional ranking on relational data, demonstrating improved generalization and computational efficiency through novel algorithms and property enforcement.

Contribution

It presents new algorithms for conditional ranking that incorporate symmetry and reciprocity, with theoretical analysis and state-of-the-art experimental results.

Findings

Ranking loss optimization outperforms regression loss in generalization.

Enforcing symmetry or reciprocity improves predictive accuracy.

Proposed methods are computationally efficient and scalable.

Abstract

In domains like bioinformatics, information retrieval and social network analysis, one can find learning tasks where the goal consists of inferring a ranking of objects, conditioned on a particular target object. We present a general kernel framework for learning conditional rankings from various types of relational data, where rankings can be conditioned on unseen data objects. We propose efficient algorithms for conditional ranking by optimizing squared regression and ranking loss functions. We show theoretically, that learning with the ranking loss is likely to generalize better than with the regression loss. Further, we prove that symmetry or reciprocity properties of relations can be efficiently enforced in the learned models. Experiments on synthetic and real-world data illustrate that the proposed methods deliver state-of-the-art performance in terms of predictive power and computational efficiency. Moreover, we also show empirically that incorporating symmetry or reciprocity properties can improve the generalization performance.