Mixup Regularized Adversarial Networks for Multi-Domain Text Classification

TL;DR

This paper introduces MRAN, a novel mixup regularized adversarial network that enhances multi-domain text classification by enriching features and improving domain invariance, achieving superior accuracy on benchmark datasets.

Contribution

The paper proposes a mixup regularization technique integrated with adversarial training to better extract domain-invariant features in multi-domain text classification.

Findings

Achieved 87.64% accuracy on Amazon review dataset.

Achieved 89.0% accuracy on FDU-MTL dataset.

Outperformed all relevant baselines on both datasets.

Abstract

Using the shared-private paradigm and adversarial training has significantly improved the performances of multi-domain text classification (MDTC) models. However, there are two issues for the existing methods. First, instances from the multiple domains are not sufficient for domain-invariant feature extraction. Second, aligning on the marginal distributions may lead to fatal mismatching. In this paper, we propose a mixup regularized adversarial network (MRAN) to address these two issues. More specifically, the domain and category mixup regularizations are introduced to enrich the intrinsic features in the shared latent space and enforce consistent predictions in-between training instances such that the learned features can be more domain-invariant and discriminative. We conduct experiments on two benchmarks: The Amazon review dataset and the FDU-MTL dataset. Our approach on these two datasets yields average accuracies of 87.64\% and 89.0\% respectively, outperforming all relevant baselines.