Few-shot fake news detection via prompt-based tuning

Abstract

 As people increasingly use social media to read news, fake news has become a major problem for the public and government. One of the main challenges in fake news detection is how to identify them in the early stage of propagation. Another challenge is that detection model training requires large amounts of labeled data, which are often unavailable or expensive to acquire. To address these challenges, we propose a novel Fake News Detection model based on Prompt Tuning (FNDPT). FNDPT first designs a prompt-based template for early fake news detection. This mechanism incorporates contextual information into textual content and extracts relevant knowledge from pre-trained language models. Furthermore, our model utilizes prompt-based tuning to enhance the performance in a few-shot setting. Experimental results on two real-world datasets verify the effectiveness of FNDPT.