Supportiveness-based Knowledge Rewriting for Retrieval-augmented Language Modeling

TL;DR

This paper proposes Supportiveness-based Knowledge Rewriting (SKR), a method to improve retrieval-augmented language models by filtering and optimizing external knowledge based on how effectively it supports downstream tasks.

Contribution

The paper introduces a novel supportiveness metric and a training strategy for knowledge rewriting, enhancing the reliability and usefulness of retrieved knowledge for LLMs.

Findings

SKR outperforms GPT-4 in knowledge rewriting with only 7B parameters.

SKR improves performance across six knowledge-intensive tasks.

The method effectively filters irrelevant knowledge and aligns rewrites with task supportiveness.

Abstract

Retrieval-augmented language models (RALMs) have recently shown great potential in mitigating the limitations of implicit knowledge in LLMs, such as untimely updating of the latest expertise and unreliable retention of long-tail knowledge. However, since the external knowledge base, as well as the retriever, can not guarantee reliability, potentially leading to the knowledge retrieved not being helpful or even misleading for LLM generation. In this paper, we introduce Supportiveness-based Knowledge Rewriting (SKR), a robust and pluggable knowledge rewriter inherently optimized for LLM generation. Specifically, we introduce the novel concept of "supportiveness"--which represents how effectively a knowledge piece facilitates downstream tasks--by considering the perplexity impact of augmented knowledge on the response text of a white-box LLM. Based on knowledge supportiveness, we first design a training data curation strategy for our rewriter model, effectively identifying and filtering out poor or irrelevant rewrites (e.g., with low supportiveness scores) to improve data efficacy. We then introduce the direct preference optimization (DPO) algorithm to align the generated rewrites to optimal supportiveness, guiding the rewriter model to summarize augmented content that better improves the final response. Comprehensive evaluations across six popular knowledge-intensive tasks and four LLMs have demonstrated the effectiveness and superiority of SKR. With only 7B parameters, SKR has shown better knowledge rewriting capability over GPT-4, the current state-of-the-art general-purpose LLM.