Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Large Language Model Agent for Fake News Detection

Xinyi Li, Yongfeng Zhang, Edward C. Malthouse

TL;DR

This work introduces FactAgent, an agentic framework that uses large language models to verify news claims without training by decomposing fake news detection into a structured workflow that combines internal knowledge with external tools. The approach employs a domain-informed suite of tools (including Phrase, Language, Commonsense, Standing, Search, and URL checks) and relies on explicit step-by-step reasoning to produce transparent final verdicts. Experimental results across Snopes, PolitiFact, and GossipCop show FactAgent outperforms traditional supervised models and prior LLM-based methods, with significant gains driven by domain knowledge and external evidence integration. The study also demonstrates the importance of the tool design and decision-making strategy, and highlights the interpretability and adaptability of FactAgent for cross-domain news verification.

Abstract

In the current digital era, the rapid spread of misinformation on online platforms presents significant challenges to societal well-being, public trust, and democratic processes, influencing critical decision making and public opinion. To address these challenges, there is a growing need for automated fake news detection mechanisms. Pre-trained large language models (LLMs) have demonstrated exceptional capabilities across various natural language processing (NLP) tasks, prompting exploration into their potential for verifying news claims. Instead of employing LLMs in a non-agentic way, where LLMs generate responses based on direct prompts in a single shot, our work introduces FactAgent, an agentic approach of utilizing LLMs for fake news detection. FactAgent enables LLMs to emulate human expert behavior in verifying news claims without any model training, following a structured workflow. This workflow breaks down the complex task of news veracity checking into multiple sub-steps, where LLMs complete simple tasks using their internal knowledge or external tools. At the final step of the workflow, LLMs integrate all findings throughout the workflow to determine the news claim's veracity. Compared to manual human verification, FactAgent offers enhanced efficiency. Experimental studies demonstrate the effectiveness of FactAgent in verifying claims without the need for any training process. Moreover, FactAgent provides transparent explanations at each step of the workflow and during final decision-making, offering insights into the reasoning process of fake news detection for end users. FactAgent is highly adaptable, allowing for straightforward updates to its tools that LLMs can leverage within the workflow, as well as updates to the workflow itself using domain knowledge. This adaptability enables FactAgent's application to news verification across various domains.

Large Language Model Agent for Fake News Detection

TL;DR

This work introduces FactAgent, an agentic framework that uses large language models to verify news claims without training by decomposing fake news detection into a structured workflow that combines internal knowledge with external tools. The approach employs a domain-informed suite of tools (including Phrase, Language, Commonsense, Standing, Search, and URL checks) and relies on explicit step-by-step reasoning to produce transparent final verdicts. Experimental results across Snopes, PolitiFact, and GossipCop show FactAgent outperforms traditional supervised models and prior LLM-based methods, with significant gains driven by domain knowledge and external evidence integration. The study also demonstrates the importance of the tool design and decision-making strategy, and highlights the interpretability and adaptability of FactAgent for cross-domain news verification.

Abstract

In the current digital era, the rapid spread of misinformation on online platforms presents significant challenges to societal well-being, public trust, and democratic processes, influencing critical decision making and public opinion. To address these challenges, there is a growing need for automated fake news detection mechanisms. Pre-trained large language models (LLMs) have demonstrated exceptional capabilities across various natural language processing (NLP) tasks, prompting exploration into their potential for verifying news claims. Instead of employing LLMs in a non-agentic way, where LLMs generate responses based on direct prompts in a single shot, our work introduces FactAgent, an agentic approach of utilizing LLMs for fake news detection. FactAgent enables LLMs to emulate human expert behavior in verifying news claims without any model training, following a structured workflow. This workflow breaks down the complex task of news veracity checking into multiple sub-steps, where LLMs complete simple tasks using their internal knowledge or external tools. At the final step of the workflow, LLMs integrate all findings throughout the workflow to determine the news claim's veracity. Compared to manual human verification, FactAgent offers enhanced efficiency. Experimental studies demonstrate the effectiveness of FactAgent in verifying claims without the need for any training process. Moreover, FactAgent provides transparent explanations at each step of the workflow and during final decision-making, offering insights into the reasoning process of fake news detection for end users. FactAgent is highly adaptable, allowing for straightforward updates to its tools that LLMs can leverage within the workflow, as well as updates to the workflow itself using domain knowledge. This adaptability enables FactAgent's application to news verification across various domains.
Paper Structure (12 sections, 8 figures, 2 tables)

This paper contains 12 sections, 8 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Experiments and Results
  5. Experimental Setup
  6. Experimental Results
  7. Fake News Detection Performance (RQ1)
  8. Importance of Domain Knowledge (RQ2)
  9. Importance of External Search Engine (RQ3)
  10. Ablation Study on Decision Making Strategy (RQ4)
  11. Case Study
  12. Conclusion

Figures (8)

  • Figure 1: The structured expert workflow for fake news detection is depicted in this diagram. The "Standing_tool" is highlighted with a dashed frame, and the fifth bullet point is shaded grey to indicate that the "Standing_tool" and its corresponding checklist item are skipped if the news is not relevant to politics. The Observations section comprises a list of observations collected from each tool sequentially. The News is represented using its title, domain URL, and publish date, formatted as 'Title: Riverdale Set to Recast a Major Character Ahead of Season 2, Domain URL: tvline.com, Publish Date: 04/25/2017'. If the domain URL and publish date are unavailable, only the title information is used.
  • Figure 2: Instructions for the LLM to automatically generate a self-designed workflow for fake news detection. The News is represented using its title, domain URL, and publish date if available.
  • Figure 3: Performance comparison of FactAgent following an automatically self-designed workflow and an expert workflow.
  • Figure 4: The frequency of each tool's usage among the testing samples when the LLM creates a self-designed workflow to evaluate news veracity.
  • Figure 5: Performance comparison of FactAgent with and without utilizing the Standing_tool within the expert workflow.
  • ...and 3 more figures