Handling Class Imbalance in Low-Resource Dialogue Systems by Combining Few-Shot Classification and Interpolation

TL;DR

This paper introduces a novel end-to-end pairwise learning framework that enhances low-resource dialogue systems by combining few-shot classification and representation interpolation, significantly improving utterance classification performance.

Contribution

The proposed method is a general training approach that induces few-shot capabilities and data augmentation through interpolation, applicable across various neural architectures.

Findings

Significant macro-F1 score improvements over standard training.

Effective across multiple neural architectures.

Demonstrated on Virtual Patient and Switchboard datasets.

Abstract

Utterance classification performance in low-resource dialogue systems is constrained by an inevitably high degree of data imbalance in class labels. We present a new end-to-end pairwise learning framework that is designed specifically to tackle this phenomenon by inducing a few-shot classification capability in the utterance representations and augmenting data through an interpolation of utterance representations. Our approach is a general purpose training methodology, agnostic to the neural architecture used for encoding utterances. We show significant improvements in macro-F1 score over standard cross-entropy training for three different neural architectures, demonstrating improvements on a Virtual Patient dialogue dataset as well as a low-resourced emulation of the Switchboard dialogue act classification dataset.