基于对抗神经网络的突发公共卫生事件虚假新闻检测

Detection of fake news on emergency public health events based on adversarial neural network

目的:在缺乏相应先验知识和训练语料的情况下,实现对突发公共卫生事件社交媒体虚假新闻的早期检测。方法:融合虚假新闻文本的语义特征和统计特征,构建基于对抗神经网络的跨领域虚假新闻检测模型,并使用新浪微博数据集开展实验。结果:基于对抗神经网络的跨领域虚假新闻检测模型的检测效果较好,检测正确率达85.6%。结论:深度语义特征与传统统计特征相结合能够更好地辅助虚假新闻的识别,对抗神经网络能够在训练过程中提取更多虚假新闻的潜在通用特征,从而提升模型的领域迁移能力,更好地进行突发公共卫生事件虚假新闻的检测。

Objective To early detect the fake news on emergency public health events in absence of prior knowledge and training corpus.Methods The model for detecting the fake news in cross domain field was established based on adversarial neural network by fusing the semantic and statistical features of fake news text.The experiment was carried out using the Sina Microblog News Dataset.Results The cross domain detection model of fake news on emergency public health events achieved rather good results with an accuracy of 85.6%.Conclusion Deep semantic characteristics combined with traditional statistical features can effectively recognize the fake news on emergency public health events,extract more potential common features of fake news on emergency public health events in the training process,improve the domain migration ability,and detect the fake news on emergency public health events.