Mitigating Hallucinations in Large Language Models via Causal Reasoning

2025-08-17

AI Summary

Large language models (LLMs) suffer from logical inconsistency and hallucinations, primarily due to their inability to model causal structures among variables; existing approaches like chain-of-thought operate only at the token level and fail to encode conditional independence or satisfy causal identifiability assumptions. Method: We propose the first variable-level causal directed acyclic graph (DAG) modeling framework, construct the CausalDR dataset (25K samples), and introduce CDCR-SFT—a supervised fine-tuning method integrating causal graph construction, graph-based reasoning tracing, and causal identifiability verification. Results: Our approach substantially improves logical consistency: it achieves 95.33% accuracy on CLADDER—surpassing human performance for the first time—and reduces hallucination rates by 10% on HaluEval. It consistently outperforms baselines across four major LLMs and eight reasoning benchmarks.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Knowledge Representation and Reasoning: Action, Change, and Causality

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

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.

Problem

• Mitigating logically inconsistent hallucinations in LLMs
• Enhancing causal reasoning capabilities in language models
• Reducing logical inconsistencies via explicit causal structure modeling

Innovation

• Introduces causal-DAG construction and reasoning framework
• Trains LLMs to model variable-level causal relationships
• Uses supervised fine-tuning to reduce hallucinations