Deliberation on Priors: Trustworthy Reasoning of Large Language Models on Knowledge Graphs

TL;DR

This paper introduces Deliberation over Priors (DP), a framework that enhances the trustworthiness of LLMs by effectively utilizing knowledge graph priors through progressive knowledge distillation and reasoning verification, leading to improved factual accuracy.

Contribution

The paper presents a novel framework that fully exploits knowledge graph priors in LLMs, combining structural integration and reasoning verification for trustworthy responses.

Findings

Achieves state-of-the-art performance with a 13% Hit@1 improvement on ComplexWebQuestions.

Enhances faithfulness of relation path generation in LLMs.

Ensures response reliability through reasoning-introspection strategies.

Abstract

Knowledge graph-based retrieval-augmented generation seeks to mitigate hallucinations in Large Language Models (LLMs) caused by insufficient or outdated knowledge. However, existing methods often fail to fully exploit the prior knowledge embedded in knowledge graphs (KGs), particularly their structural information and explicit or implicit constraints. The former can enhance the faithfulness of LLMs' reasoning, while the latter can improve the reliability of response generation. Motivated by these, we propose a trustworthy reasoning framework, termed Deliberation over Priors (DP), which sufficiently utilizes the priors contained in KGs. Specifically, DP adopts a progressive knowledge distillation strategy that integrates structural priors into LLMs through a combination of supervised fine-tuning and Kahneman-Tversky optimization, thereby improving the faithfulness of relation path generation. Furthermore, our framework employs a reasoning-introspection strategy, which guides LLMs to perform refined reasoning verification based on extracted constraint priors, ensuring the reliability of response generation. Extensive experiments on three benchmark datasets demonstrate that DP achieves new state-of-the-art performance, especially a Hit@1 improvement of 13% on the ComplexWebQuestions dataset, and generates highly trustworthy responses. We also conduct various analyses to verify its flexibility and practicality. The code is available at https://github.com/reml-group/Deliberation-on-Priors.