Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Explore the Potential of LLMs in Misinformation Detection: An Empirical Study

Mengyang Chen, Lingwei Wei, Han Cao, Wei Zhou, Songlin Hu

TL;DR

This paper systematically investigates the potential of large language models for misinformation detection, separating evaluation into LLM-based detectors (direct prompting) and LLM-enhanced detectors (embedding and data augmentation). It demonstrates that LLMs can match or exceed small models on text-based misinformation when prompted effectively, but struggle with propagation-aware detection due to graph-structured data. The study shows that LLM-enhanced detectors, using LLM embeddings or generated data, often improve performance over traditional baselines, with notable gains from Chinese LLMs on Chinese datasets. Together, these findings highlight the practical value and limitations of deploying LLMs for misinformation detection and point to promising directions for integrating LLMs with existing detectors.

Abstract

Large Language Models (LLMs) have garnered significant attention for their powerful ability in natural language understanding and reasoning. In this paper, we present a comprehensive empirical study to explore the performance of LLMs on misinformation detection tasks. This study stands as the pioneering investigation into the understanding capabilities of multiple LLMs regarding both content and propagation across social media platforms. Our empirical studies on eight misinformation detection datasets show that LLM-based detectors can achieve comparable performance in text-based misinformation detection but exhibit notably constrained capabilities in comprehending propagation structure compared to existing models in propagation-based misinformation detection. Our experiments further demonstrate that LLMs exhibit great potential to enhance existing misinformation detection models. These findings highlight the potential ability of LLMs to detect misinformation.

Explore the Potential of LLMs in Misinformation Detection: An Empirical Study

TL;DR

This paper systematically investigates the potential of large language models for misinformation detection, separating evaluation into LLM-based detectors (direct prompting) and LLM-enhanced detectors (embedding and data augmentation). It demonstrates that LLMs can match or exceed small models on text-based misinformation when prompted effectively, but struggle with propagation-aware detection due to graph-structured data. The study shows that LLM-enhanced detectors, using LLM embeddings or generated data, often improve performance over traditional baselines, with notable gains from Chinese LLMs on Chinese datasets. Together, these findings highlight the practical value and limitations of deploying LLMs for misinformation detection and point to promising directions for integrating LLMs with existing detectors.

Abstract

Large Language Models (LLMs) have garnered significant attention for their powerful ability in natural language understanding and reasoning. In this paper, we present a comprehensive empirical study to explore the performance of LLMs on misinformation detection tasks. This study stands as the pioneering investigation into the understanding capabilities of multiple LLMs regarding both content and propagation across social media platforms. Our empirical studies on eight misinformation detection datasets show that LLM-based detectors can achieve comparable performance in text-based misinformation detection but exhibit notably constrained capabilities in comprehending propagation structure compared to existing models in propagation-based misinformation detection. Our experiments further demonstrate that LLMs exhibit great potential to enhance existing misinformation detection models. These findings highlight the potential ability of LLMs to detect misinformation.
Paper Structure (29 sections, 5 figures, 11 tables)

This paper contains 29 sections, 5 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Misinformation Detection
  4. Large Language Models
  5. Preminarlies
  6. Misinformation Detection
  7. LLM-based Detectors
  8. LLM-enhanced Detectors
  9. Evaluation of LLM-based Detectors
  10. Evaluation Settings
  11. Datasets
  12. Comparison Methods
  13. Selection of SLMs
  14. Selection of LLMs
  15. Prompt Settings
  16. ...and 14 more sections

Figures (5)

  • Figure 1: Two paradigms for utilizing LLMs in misinformation detection.
  • Figure 2: Few-shot Learning results on three text-based misinformation detection datasets.
  • Figure 3: Accuracy scores (%) of instruction-tuning LLaMa2-7B against different training data size.
  • Figure 4: Changes in news authenticity lead to LLMs' misjudgment (Taking GPT-3.5-turbo as an example).
  • Figure 5: Results (%) of LLMs against different categories of news.