基于对抗图增强对比学习的虚假新闻检测

Fake News Detection Based on Adversarial Graph Enhanced Contrastive Learning

随着互联网的快速发展,社交媒体成为了新闻发布和传播的主要平台,如何准确识别虚假新闻已成为研究热点。现有的基于深度学习的虚假新闻检测方法在面对噪声和敌对信息时缺乏鲁棒性。为了应对这一挑战,该文提出了一种对抗图增强对比学习的方法,该方法引入对抗对比学习,使模型抓住少量但充分的信息完成增强图与原始图之间的互信息最大化,在进行训练时重点捕捉有用信息。同时,该模型还利用了特征增强器和图表示对比学习进行图表示增强,加强特征学习。在两个公共数据集上进行的实验表明,该模型在现有基线上达到了最优的性能。

Existing deep learning-based fake news detection methods are defected in dealing with face of noise and hostile information.This paper proposes an adversarial graph-enhanced contrastive learning method to address this issue.It introduces adversarial contrastive learning so that the model can maximize the mutual information between the enhanced graph and the original graph.Specifically,this model also applies a feature enhancer to the graph representation contrastive learning.Experimental results on two public datasets show that the proposed model achieves state-of-the-art performance.