Beyond Perplexity: Let the Reader Select Retrieval Summaries via Spectrum Projection Score

TL;DR

This paper introduces Spectrum Projection Score (SPS), a novel metric for evaluating retrieval relevance in LLMs, and presents xCompress, a framework that improves retrieval-augmented generation by dynamically selecting and compressing summaries, leading to better QA performance.

Contribution

The paper proposes SPS as a supervision-free metric for assessing retrieval relevance and introduces xCompress, an inference-time controller for improved retrieval-augmented generation.

Findings

SPS enhances QA performance across multiple benchmarks.

xCompress improves the selection and compression of retrieval summaries.

The approach provides a new perspective on retrieval and generation interaction.

Abstract

Large Language Models (LLMs) have shown improved generation performance through retrieval-augmented generation (RAG) following the retriever-reader paradigm, which supplements model inputs with externally retrieved knowledge. However, prior work often evaluates RAG holistically, assessing the retriever and reader jointly, making it difficult to isolate the true contribution of retrieval, particularly given the prompt sensitivity of LLMs used as readers. We move beyond perplexity and introduce Spectrum Projection Score (SPS), a lightweight and supervision-free metric that allows the reader to gauge the semantic alignment of a retrieved summary with its hidden representation by comparing the area formed by generated tokens from the summary, and the principal directions of subspace in the reader and to measure the relevance. Building on SPS we present xCompress, an inference-time controller framework that dynamically samples, ranks, and compresses retrieval summary candidates. Extensive experiments on five QA benchmarks with four open-sourced LLMs show that SPS not only enhances performance across a range of tasks but also provides a principled perspective on the interaction between retrieval and generation.