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

Dongxu Zhang, Varun Gangal, Barrett Martin Lattimer, Yi Yang
7/7/2024
cs.AIcs.CL

No official code found. Best implementation: asappresearch/halugen (1 star, MIT, last commit 2024-07-07).

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outputs is pivotal, yet traditional fine-tuning for this classification task is impeded by the expensive and quickly outdated annotation process, especially across numerous vertical domains and in the face of rapid LLM advancements.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.

View on arXivView PDF

Code Implementations(1)

asappresearch/halugenCommunity100%
MIT

https://openreview.net/pdf?id=T1kZ0tdOtZM

10PythonJun 5, 20242 years agoMIT

Similar Papers

Unsupervised Real-Time Hallucination Detection based on the Internal States of Large Language Models

92% match1 repo652024

A novel hallucination classification framework

92% match2025

From Detection to Diagnosis: Advancing Hallucination Analysis with Automated Data Synthesis

92% match2025

Hallucination Detection with the Internal Layers of LLMs

91% match2025

(Im)possibility of Automated Hallucination Detection in Large Language Models

91% match2025

Fine-grained Hallucination Detection and Editing for Language Models

91% match2024

Cite this paper

@article{zhang2024enhancing,
  title  = {Enhancing Hallucination Detection through Perturbation-Based Synthetic Data Generation in System Responses},
  author = {Dongxu Zhang and Varun Gangal and Barrett Martin Lattimer and Yi Yang},
  year   = {2024},
  eprint = {2407.05474},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2407.05474v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2407.05474.svg)](https://codeofpaper.com/paper/2407.05474)

Discussion