基于BERT-BiLSTM模型的虚假新闻检测  

False News Detection Based on Bert-BiLSTM Model

作　　者：张敏超 蒲秋梅 黄方俐[1,2] ZHANG Min-chao;PU Qiu-mei;HUANG Fang-li(Key Laboratory of Ethnic Language Intelligent Analysis and Security Governance of MOE,Minzu University of China,Beijing 100081,China;School of Information Engineering,Minzu University of China,Beijing 100081,China)

机构地区：[1]民族语言智能分析与安全治理教育部重点实验室,北京100081 [2]中央民族大学信息工程学院,北京100081

出　　处：《中国电子科学研究院学报》2025年第1期33-40,共8页Journal of China Academy of Electronics and Information Technology

基　　金：国家社会科学基金项目(20BGL251)。

摘　　要：随着互联网的快速发展,虚假新闻的传播成为全球性问题,严重影响社会稳定和信息安全,因此,如何有效识别虚假新闻已成为自然语言处理领域的研究重点之一。文中提出了一种基于BERT的虚假新闻检测模型。该模型首先通过对预训练的BERT模型进行微调,以获取新闻文本的深层语义表示;然后,在其顶部分别添加BiLSTM层和卷积神经网络(Convolutional Neural Networks,CNN)层,以捕捉文本的长程依赖关系和局部上下文特征;最后,通过Softmax层实现虚假新闻的分类。实验在新闻文本数据集上进行,结果显示,BERT-BiLSTM模型在英文数据集上的准确率达到96.14%,在中文数据集上的准确率达到97.32%。相比其他模型,BERT-BiLSTM在虚假新闻检测中表现更为优异,具有良好的实际应用价值,对维护网络信息安全具有重要意义。With the rapid development of the Internet,the spread of false news has become a global problem,seriously affecting social stability and information security.Therefore,how to effectively identify false news has become one of the research focuses in the field of natural language processing.This paper proposes a false news detection model based on BERT.The model begins by fine-tuning a pretrained BERT model to obtain a deep semantic representation of the news text,and then adds BiLSTM and CNN layers on top of it,respectively,to capture the long-range dependencies and local contextual features of the text.Finally,the classification of fake news is realized through the Softmax layer.The results show that the accuracy of BERT-BiLSTM model is 96.14%on English data set and 97.32%on Chinese data set.Compared with other models,BERT-BiLSTM performs better in false news detection,has good practical application value,and is of great significance for maintaining network information security.

关 键 词：BERT模型 BiLSTM模型 虚假新闻检测 深度学习 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]