Improving Contrastive Learning of Sentence Embeddings with Case-Augmented Positives and Retrieved Negatives

TL;DR

This paper introduces CARDS, a contrastive learning method that enhances sentence embeddings by using case-augmented positives and retrieved hard negatives, achieving state-of-the-art results in unsupervised settings.

Contribution

It proposes novel case augmentation and hard negative sampling techniques to improve the quality of contrastive learning for sentence embeddings.

Findings

CARDS outperforms previous SOTA methods on STS benchmarks

Case augmentation reduces bias in token embeddings

Hard negative sampling improves embedding discrimination

Abstract

Following SimCSE, contrastive learning based methods have achieved the state-of-the-art (SOTA) performance in learning sentence embeddings. However, the unsupervised contrastive learning methods still lag far behind the supervised counterparts. We attribute this to the quality of positive and negative samples, and aim to improve both. Specifically, for positive samples, we propose switch-case augmentation to flip the case of the first letter of randomly selected words in a sentence. This is to counteract the intrinsic bias of pre-trained token embeddings to frequency, word cases and subwords. For negative samples, we sample hard negatives from the whole dataset based on a pre-trained language model. Combining the above two methods with SimCSE, our proposed Contrastive learning with Augmented and Retrieved Data for Sentence embedding (CARDS) method significantly surpasses the current SOTA on STS benchmarks in the unsupervised setting.