[论文解读] FakeGPT: Fake News Generation, Explanation and Detection of Large Language Models

Yue Huang, Lichao Sun|arXiv (Cornell University)|Oct 8, 2023

Misinformation and Its Impacts被引用 11

一句话总结

该论文研究了 ChatGPT 在生成、解释和检测假新闻方面的能力，并引入了一种具备推理意识的提示方法以提升检测性能。

ABSTRACT

The rampant spread of fake news has adversely affected society, resulting in extensive research on curbing its spread. As a notable milestone in large language models (LLMs), ChatGPT has gained significant attention due to its exceptional natural language processing capabilities. In this study, we present a thorough exploration of ChatGPT's proficiency in generating, explaining, and detecting fake news as follows. Generation -- We employ four prompt methods to generate fake news samples and prove the high quality of these samples through both self-assessment and human evaluation. Explanation -- We obtain nine features to characterize fake news based on ChatGPT's explanations and analyze the distribution of these factors across multiple public datasets. Detection -- We examine ChatGPT's capacity to identify fake news. We explore its detection consistency and then propose a reason-aware prompt method to improve its performance. Although our experiments demonstrate that ChatGPT shows commendable performance in detecting fake news, there is still room for its improvement. Consequently, we further probe into the potential extra information that could bolster its effectiveness in detecting fake news.

研究动机与目标

• 评估 ChatGPT 使用多种提示策略在自评与人评两种评估下生成假新闻的能力。
• 识别并分类解释假新闻为何具有欺骗性的特征。
• 评估 ChatGPT 的假新闻检测表现，并开发基于提示的改进方法。
• 探讨可能进一步提升检测效果的附加信息。

提出的方法

• 四种提示方法用于生成假新闻进行测试，并对样本质量进行自评与他评。
• 对 ChatGPT 进行解释假新闻，并从解释中提取九个定义性特征。
• 提出一种具备推理意识的提示策略，以提高 ChatGPT 的假新闻检测准确性。
• 实验包含多个数据集，以评估检测的一致性以及附加信息（如上下文）的影响。
• 评估包括 2 类（假/真）与 3 类（假/真/不确定）的设置，并使用专门的度量指标。

Figure 1: Multiple prompts for fake news generation through ChatGPT. The words in red mean details of generated fake news.

实验结果

研究问题

• RQ1 Can ChatGPT generate high-quality fake-news samples across different prompting strategies?
• RQ2 What nine features best characterize fake news from ChatGPT explanations?
• RQ3 Does a reason-aware prompt improve ChatGPT’s ability to detect fake news across datasets?
• RQ4 What additional information (e.g., context, knowledge) aids fake-news detection with ChatGPT?

主要发现

• ChatGPT 可以生成高质量的假新闻样本，在自评和他评上都可与真实新闻相比。
• 九个特征在来自 ChatGPT 的解释中识别出假新闻，在九个数据集中成立。
• 具备推理意识的提示在大多数数据集上提升了 ChatGPT 的假新闻检测准确性，尤其是在两类任务中。
• 额外信息如上下文和评论通常能提升检测，尽管效果因数据集与任务设置而异。
• ChatGPT 在某些数据集上的检测表现强劲，但在样本和数据集之间仍不完善且不一致。
• 在三类别设置中，某数据集的最高报告准确率为 82.6%，且由具备推理意识提示带来显著提升。

Figure 2: Four kinds of the prompt template.