基于多维度行文风格特征的假新闻检测方法研究

Research on Fake News Detection Method based on Multi-dimensional Writing Style Features

传统的基于行文风格的假新闻检测方法对文本的行文风格特征考虑不全面,只考虑符号、句长、句数和特征词,而忽略情绪词、虚词、实词等行文风格特征,并且不能很好地提取真假新闻行文风格之间的差异性信息,因此探索一种基于词、句、篇三个维度的行文风格特征,用于自动检测社交媒体上低可信度文本的方法。充分研究文本多维度行文风格特征对假新闻检测的作用,使用多通道卷积神经网络提取行文风格的高阶抽象信息,利用注意力机制捕获各维度特征对假新闻检测的影响力。实验显示,提出的Multi_CNNSA模型显著提高了假新闻检测效果,在weibo数据集上取得86.95%的F 1值。

Traditional fake news detection methods based on writing style do not fully consider the writing style characteristics of the text.They only consider symbols,sentence length,sentence number and feature words,while ignoring the writing style features such as emotional words,functional words,and substantial words.They are not good enough to extract difference information between the writing styles of true and fake news.In view of these problems,this paper proposes to explore a method for automatic detection of low-credibility texts on social media based on the three-dimensional writing style features of words,sentences,and contexts.After a full study on the effect of text multi-dimensional style features on fake news detection,multi-channel convolutional neural network is used to extract high-level abstract information of style,and attention mechanism is used to capture the influence of each dimension feature on fake news detection.Experiments show that the proposed Multi_CNNSA model significantly improves the effect of fake news detection,achieving an F 1 value of 86.95%on the weibo dataset.