GASE: Generatively Augmented Sentence Encoding

TL;DR

GASE introduces a training-free method that enhances sentence embeddings at inference time by using generative models for data augmentation, improving performance without model fine-tuning.

Contribution

It presents a novel inference-time augmentation technique using generative models to improve sentence embeddings without requiring parameter access or additional training.

Findings

Performance improves across various embedding models.

Greater gains observed for models with lower baseline performance.

Enhances robustness and semantic diversity of sentence embeddings.

Abstract

We propose a training-free approach to improve sentence embeddings leveraging test-time compute by applying generative text models for data augmentation at inference time. Unlike conventional data augmentation that utilises synthetic training data, our approach does not require access to model parameters or the computational resources typically required for fine-tuning state-of-the-art models. Generatively Augmented Sentence Encoding variates the input text by paraphrasing, summarising, or extracting keywords, followed by pooling the original and synthetic embeddings. Experimental results on the Massive Text Embedding Benchmark for Semantic Textual Similarity (STS) demonstrate performance improvements across a range of embedding models using different generative models for augmentation. We find that generative augmentation leads to larger performance improvements for embedding models with lower baseline performance. These findings suggest that integrating generative augmentation at inference time adds semantic diversity and can enhance the robustness and generalisability of sentence embeddings for embedding models. Our results show that performance gains depend on the embedding model and the dataset.