基于事实和风格的多通道融合虚假新闻检测

Multi-channel Fusion of Style and Fact for Fake News Detection

虚假新闻的大量传播对个人和社会都造成巨大的危害,通过智能算法检测虚假新闻是阻止虚假新闻传播的重要途径。针对不同语境中虚假新闻检测不准确的问题,该文将新闻的背景事实特征和新闻的风格特征融入到模型中,可以提高模型解决缺少背景知识的虚假新闻检测能力,增强模型的鲁棒性,其中新闻的风格包括情感风格和文本风格。同时该文构建了多通道融合器融合新闻与背景知识的差异性特征,语义特征和风格特征,组成了基于事实和风格的虚假新闻检测框架FSFD。在CHEF中文开放数据集上的实验证明,该文提出的检测方法在F1值上比基准模型提升了2.3%,可见,该文方法适用于背景丰富的新闻,为在线社交媒体的虚假新闻检测提供有力支持。

The widespread dissemination of fake news poses significant harm to individuals and society.To detect fake news in different contexts,this paper incorporates the background factual features and stylistic features of news into the model.This integration enhances the model's ability to detect fake news lacking background knowledge and improves its robustness.The stylistic features include emotional style and textual style.Additionally,this paper constructs a multi-channel integrator that combines differential features of news and background knowledge,semantic features,and stylistic features,forming the FSFD framework.Validation using the CHEF Chinese open dataset demonstrates that the proposed method outperforms the baseline model by 2.3%in terms of F 1 score.