Enhancing Hallucination Detection through Perturbation-Based Synthetic Data Generation in System Responses

TL;DR

This paper presents a perturbation-based synthetic data generation method to improve hallucination detection in LLM outputs, outperforming existing zero-shot detectors and synthetic methods in accuracy and speed.

Contribution

The study introduces an automatic response rewriting approach to generate training data, enhancing hallucination detection without costly manual annotation.

Findings

T5-base model fine-tuned on generated data outperforms state-of-the-art zero-shot detectors.

The approach improves detection accuracy and reduces latency.

Synthetic data generation effectively adapts to rapid LLM advancements.

Abstract

Detecting hallucinations in large language model (LLM) outputs is pivotal, yet traditional fine-tuning for this classification task is impeded by the expensive and quickly outdated annotation process, especially across numerous vertical domains and in the face of rapid LLM advancements. In this study, we introduce an approach that automatically generates both faithful and hallucinated outputs by rewriting system responses. Experimental findings demonstrate that a T5-base model, fine-tuned on our generated dataset, surpasses state-of-the-art zero-shot detectors and existing synthetic generation methods in both accuracy and latency, indicating efficacy of our approach.