Exploring the Impact of Negative Samples of Contrastive Learning: A Case Study of Sentence Embedding

TL;DR

This paper investigates how the size of negative sample queues affects contrastive learning for sentence embedding, proposing MoCoSE with mechanisms to optimize the use of historical negative samples and demonstrating improved performance on semantic similarity tasks.

Contribution

It introduces MoCoSE, a momentum contrastive learning model with a novel maximum traceable distance metric to optimize negative sample utilization in sentence embedding.

Findings

Optimal negative sample queue range improves performance.

Maximum traceable distance correlates with better embeddings.

Achieved 77.27% Spearman's correlation on STS task.

Abstract

Contrastive learning is emerging as a powerful technique for extracting knowledge from unlabeled data. This technique requires a balanced mixture of two ingredients: positive (similar) and negative (dissimilar) samples. This is typically achieved by maintaining a queue of negative samples during training. Prior works in the area typically uses a fixed-length negative sample queue, but how the negative sample size affects the model performance remains unclear. The opaque impact of the number of negative samples on performance when employing contrastive learning aroused our in-depth exploration. This paper presents a momentum contrastive learning model with negative sample queue for sentence embedding, namely MoCoSE. We add the prediction layer to the online branch to make the model asymmetric and together with EMA update mechanism of the target branch to prevent the model from collapsing. We define a maximum traceable distance metric, through which we learn to what extent the text contrastive learning benefits from the historical information of negative samples. Our experiments find that the best results are obtained when the maximum traceable distance is at a certain range, demonstrating that there is an optimal range of historical information for a negative sample queue. We evaluate the proposed unsupervised MoCoSE on the semantic text similarity (STS) task and obtain an average Spearman's correlation of $77.27\%$. Source code is available at https://github.com/xbdxwyh/mocose.