基于XLNet和循环神经网络模型的虚假信息检测研究

Research on Fake Information Detection Based on XLNet and Recurrent Neural Network Model

虚假信息借助迅速发展的社交媒体在网络上广泛传播,因此高效并准确地完成虚假信息检测任务已成为近年来自然语言处理领域的研究热点之一。现有的虚假信息检测方法存在数据训练不够准确和模型未突出关键特征影响力的问题。针对该问题,论文提出一种基于XLNet和循环神经网络模型的虚假信息检测方法。该方法基于XLNet模型对文本进行编码及特征提取,结合双向GRU模型进一步捕获文本深层语义特征,同时引入注意力机制根据词语的重要程度为文本中不同特征分别赋予不同的权重值,最后将文本的完整语义特征输出分类,实现虚假信息检测。实验结果表明,该方法在微博公开数据集和COVID-19 Fake News数据集上分别达到了94.6%和96.3%的准确率,可以有效辨别虚假信息,对于虚假信息检测任务具有一定指导意义。

Fake information is widely spread on the Internet with the help of rapidly developing social media,so efficiently and accurately completing the task of fake information detection has become one of the research hotspots in the field of natural lan-guage processing in recent years.The existing fake information detection methods have the problems that the data training is not ac-curate enough and the model does not highlight the influence of key features.Aiming at this problem,this paper proposes a fake in-formation detection method based on XLNet and recurrent neural network model.This method encodes and extracts features based on the XLNet model,and combines the bidirectional GRU model to further capture the deep semantic features of the text.At the same time,an attention mechanism is introduced to assign different weights to different features in the text according to the impor-tance of the words.Complete semantic feature output of the text is classified for fake information detection.The experimental results show that the method achieves 94.6%and 96.3%accuracy on the Weibo public dataset and the COVID-19 Fake News dataset re-spectively,which can effectively identify fake information,and has certain guiding significance for the task of fake information de-tection.