Understanding LLM Embeddings for Regression

TL;DR

This paper investigates the use of large language model embeddings as features for regression tasks, showing they can outperform traditional features and analyzing factors affecting their effectiveness.

Contribution

It provides one of the first comprehensive studies on embedding-based regression, highlighting the conditions under which LLM embeddings improve performance.

Findings

LLM embeddings can outperform traditional features in high-dimensional regression.

Embedding quality is influenced by model size and language understanding.

Lipschitz continuity in embeddings explains some performance benefits.

Abstract

With the rise of large language models (LLMs) for flexibly processing information as strings, a natural application is regression, specifically by preprocessing string representations into LLM embeddings as downstream features for metric prediction. In this paper, we provide one of the first comprehensive investigations into embedding-based regression and demonstrate that LLM embeddings as features can be better for high-dimensional regression tasks than using traditional feature engineering. This regression performance can be explained in part due to LLM embeddings over numeric data inherently preserving Lipschitz continuity over the feature space. Furthermore, we quantify the contribution of different model effects, most notably model size and language understanding, which we find surprisingly do not always improve regression performance.