Mitigating Hallucinations in Large Language Models via Causal Reasoning

TL;DR

This paper introduces a supervised fine-tuning framework called CDCR-SFT that trains large language models to explicitly construct and reason over causal DAGs, significantly reducing hallucinations and improving reasoning accuracy.

Contribution

The paper presents a novel causal-DAG construction and reasoning framework for LLMs, enabling explicit causal modeling to mitigate hallucinations and enhance reasoning performance.

Findings

Achieved 95.33% accuracy on CLADDER, surpassing human performance.

Reduced hallucinations on HaluEval by 10%.

Improved causal reasoning capabilities across multiple tasks.

Abstract

Large language models (LLMs) exhibit logically inconsistent hallucinations that appear coherent yet violate reasoning principles, with recent research suggesting an inverse relationship between causal reasoning capabilities and such hallucinations. However, existing reasoning approaches in LLMs, such as Chain-of-Thought (CoT) and its graph-based variants, operate at the linguistic token level rather than modeling the underlying causal relationships between variables, lacking the ability to represent conditional independencies or satisfy causal identification assumptions. To bridge this gap, we introduce causal-DAG construction and reasoning (CDCR-SFT), a supervised fine-tuning framework that trains LLMs to explicitly construct variable-level directed acyclic graph (DAG) and then perform reasoning over it. Moreover, we present a dataset comprising 25,368 samples (CausalDR), where each sample includes an input question, explicit causal DAG, graph-based reasoning trace, and validated answer. Experiments on four LLMs across eight tasks show that CDCR-SFT improves the causal reasoning capability with the state-of-the-art 95.33% accuracy on CLADDER (surpassing human performance of 94.8% for the first time) and reduces the hallucination on HaluEval with 10% improvements. It demonstrates that explicit causal structure modeling in LLMs can effectively mitigate logical inconsistencies in LLM outputs. Code is available at https://github.com/MrLYG/CDCR-SFT.