HalluClean: A Unified Framework to Combat Hallucinations in LLMs

TL;DR

HalluClean is a lightweight, task-agnostic framework that detects and corrects hallucinations in LLM outputs through a reasoning-enhanced, multi-stage process, improving factual accuracy across diverse NLP tasks.

Contribution

It introduces a novel, zero-shot, reasoning-based approach for hallucination correction in LLMs that does not rely on external knowledge or supervised detectors.

Findings

Significantly improves factual consistency in LLM outputs.

Outperforms existing baselines across five NLP tasks.

Demonstrates effective zero-shot generalization without external knowledge.

Abstract

Large language models (LLMs) have achieved impressive performance across a wide range of natural language processing tasks, yet they often produce hallucinated content that undermines factual reliability. To address this challenge, we introduce HalluClean, a lightweight and task-agnostic framework for detecting and correcting hallucinations in LLM-generated text. HalluClean adopts a reasoning-enhanced paradigm, explicitly decomposing the process into planning, execution, and revision stages to identify and refine unsupported claims. It employs minimal task-routing prompts to enable zero-shot generalization across diverse domains, without relying on external knowledge sources or supervised detectors. We conduct extensive evaluations on five representative tasks-question answering, dialogue, summarization, math word problems, and contradiction detection. Experimental results show that HalluClean significantly improves factual consistency and outperforms competitive baselines, demonstrating its potential to enhance the trustworthiness of LLM outputs in real-world applications.