基于情感的多头注意力卷积Transformer + CNN的假新闻检测

Fake News Detection Based on Emotional Multi-Head Attention Convolution Transformer + CNN

网络上虚假新闻的传播是我们社会的主要问题之一。虚假新闻是为了误导读者,引发读者的强烈情绪,并试图在社交网络上传播,因此自动检测虚假新闻是一项艰巨的任务。尽管最近的研究已经探索了虚假新闻的不同语言模式,但情感信号的作用尚未进行深入的探讨。本文研究了情感信号在假新闻检测中的作用,利用了新闻发布者所传达的情感信息来对假新闻进行检测。我们提出了一个加入情感特征的多头注意力卷积Transformer + CNN模型(简称SMCT模型),该模型不仅能捕获文本内容的局部和全局的依赖关系,而且还能将文本内容与新闻发布者所传达的情感信息相融合来检测新闻的真假性。在两个真实的数据集上进行了大量的实验研究,我们发现将情感特征引入模型中可以明显提升在区分真假新闻方面的准确率,这一优化策略的应用有效地增强了模型在判别新闻真实性方面的能力。

The spread of false news on the network is one of the main problems in our society. Fake news is de-signed to mislead readers, arouse strong emotions among readers, and try to spread it on social networks, so the automatic detection of false news is a difficult task. Although recent studies have explored different linguistic patterns of fake news, the role of emotional signals has not been explored in depth. This paper studies the role of emotional signals in the detection of fake news, using the emotional information conveyed by news publishers to detect fake news. We propose a multi-head attention convolution Transformer + CNN model (SMCT model), which can not only capture the local and global dependence of the text content, but also integrate the text content with the emotional information conveyed by the news publisher to detect the truth and falsehood of news. After conducting a large number of experimental studies on two real datasets, we find that introducing emotional features into the model can significantly improve the accuracy in distinguishing true and false news. The application of this optimization strategy effectively enhances the ability of the model in judging the authenticity of news.