Text Encoders Lack Knowledge: Leveraging Generative LLMs for Domain-Specific Semantic Textual Similarity

TL;DR

This paper demonstrates that generative large language models (LLMs) significantly outperform encoder-based models in domain-specific semantic textual similarity tasks, especially when world knowledge is essential, by framing STS as a text generation problem.

Contribution

It introduces a novel approach of casting semantic textual similarity as a text generation task and shows generative LLMs outperform existing models on complex, knowledge-dependent benchmarks.

Findings

Generative LLMs outperform encoder-based models by 22.3% on average in domain-specific STS tasks.

New challenge sets from social media data test models' ability to use world knowledge.

Prompting strategies enhance generative LLMs' performance in complex semantic similarity tasks.

Abstract

Amidst the sharp rise in the evaluation of large language models (LLMs) on various tasks, we find that semantic textual similarity (STS) has been under-explored. In this study, we show that STS can be cast as a text generation problem while maintaining strong performance on multiple STS benchmarks. Additionally, we show generative LLMs significantly outperform existing encoder-based STS models when characterizing the semantic similarity between two texts with complex semantic relationships dependent on world knowledge. We validate this claim by evaluating both generative LLMs and existing encoder-based STS models on three newly collected STS challenge sets which require world knowledge in the domains of Health, Politics, and Sports. All newly collected data is sourced from social media content posted after May 2023 to ensure the performance of closed-source models like ChatGPT cannot be credited to memorization. Our results show that, on average, generative LLMs outperform the best encoder-only baselines by an average of 22.3% on STS tasks requiring world knowledge. Our results suggest generative language models with STS-specific prompting strategies achieve state-of-the-art performance in complex, domain-specific STS tasks.