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

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outputs is pivotal, yet traditional fine-tuning for this classification task is impeded by the expensive and quickly outdated annotation process, especially across numerous vertical domains and in the face of rapid LLM advancements.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-theart zero-shot detectors and existing synthetic generation methods in both accuracy and latency, indicating efficacy of our approach. Hallucination data LLMs used in the data F1 MNBM ('20) GPT, Bert, Rnn, ConvNet 0.780 FRANK ('21) PointerNet, bertS2S, Bart 0.694 Seahorse (early '23) T5, MT5, PALM 0.576 ScreenEval (late '23) GPT-4, longformer 0.130