SSMix: Saliency-Based Span Mixup for Text Classification

TL;DR

SSMix introduces a novel span-based mixup technique for text classification that synthesizes new sentences by combining salient spans from original texts, improving data augmentation effectiveness in NLP tasks.

Contribution

The paper proposes SSMix, a new input-level mixup method using saliency-guided span mixing, which outperforms previous hidden-level mixup approaches in NLP classification tasks.

Findings

Outperforms hidden-level mixup methods on various benchmarks.

Effective across multiple NLP classification tasks.

Preserves local information and salient tokens during augmentation.

Abstract

Data augmentation with mixup has shown to be effective on various computer vision tasks. Despite its great success, there has been a hurdle to apply mixup to NLP tasks since text consists of discrete tokens with variable length. In this work, we propose SSMix, a novel mixup method where the operation is performed on input text rather than on hidden vectors like previous approaches. SSMix synthesizes a sentence while preserving the locality of two original texts by span-based mixing and keeping more tokens related to the prediction relying on saliency information. With extensive experiments, we empirically validate that our method outperforms hidden-level mixup methods on a wide range of text classification benchmarks, including textual entailment, sentiment classification, and question-type classification. Our code is available at https://github.com/clovaai/ssmix.