Affinity Weighted Embedding

TL;DR

This paper introduces a new class of affinity weighted embedding models that iteratively reweight features and labels to improve performance over traditional linear embedding models like Wsabie and PSI.

Contribution

It proposes an iterative reweighting approach for embedding models, enhancing their ability to leverage additional data and reduce underfitting.

Findings

Initial results show improved ranking performance.

Variants of the model demonstrate flexibility.

Reweighting enhances data utilization.

Abstract

Supervised (linear) embedding models like Wsabie and PSI have proven successful at ranking, recommendation and annotation tasks. However, despite being scalable to large datasets they do not take full advantage of the extra data due to their linear nature, and typically underfit. We propose a new class of models which aim to provide improved performance while retaining many of the benefits of the existing class of embedding models. Our new approach works by iteratively learning a linear embedding model where the next iteration's features and labels are reweighted as a function of the previous iteration. We describe several variants of the family, and give some initial results.