K Venkatachalam, Al-Onazi Badriyya B, Simic Vladimir, Tirkolaee Erfan Babaee, Jana Chiranjibe
Department of Applied Cybernetics, University of Hradec Králové, Hradec Kralove, Czech Republic.
Department of Language Preparation, Princess Nourah Bint Abdulrahman University, Riyadh, Saudi Arabia.
PeerJ Comput Sci. 2023 Dec 7;9:e1666. doi: 10.7717/peerj-cs.1666. eCollection 2023.
Early identification of false news is now essential to save lives from the dangers posed by its spread. People keep sharing false information even after it has been debunked. Those responsible for spreading misleading information in the first place should face the consequences, not the victims of their actions. Understanding how misinformation travels and how to stop it is an absolute need for society and government. Consequently, the necessity to identify false news from genuine stories has emerged with the rise of these social media platforms. One of the tough issues of conventional methodologies is identifying false news. In recent years, neural network models' performance has surpassed that of classic machine learning approaches because of their superior feature extraction. This research presents Deep learning-based Fake News Detection (DeepFND). This technique has Visual Geometry Group 19 (VGG-19) and Bidirectional Long Short Term Memory (Bi-LSTM) ensemble models for identifying misinformation spread through social media. This system uses an ensemble deep learning (DL) strategy to extract characteristics from the article's text and photos. The joint feature extractor and the attention modules are used with an ensemble approach, including pre-training and fine-tuning phases. In this article, we utilized a unique customized loss function. In this research, we look at methods for detecting bogus news on the internet without human intervention. We used the Weibo, liar, PHEME, fake and real news, and Buzzfeed datasets to analyze fake and real news. Multiple methods for identifying fake news are compared and contrasted. Precision procedures have been used to calculate the proposed model's output. The model's 99.88% accuracy is better than expected.
如今，尽早识别虚假新闻对于避免其传播带来的危险、拯救生命至关重要。即便虚假信息已被揭穿，人们仍在不断分享。首先传播误导性信息的人应承担后果，而非其行为的受害者。了解错误信息的传播方式以及如何阻止它，是社会和政府的绝对需求。因此，随着这些社交媒体平台的兴起，从真实新闻中识别虚假新闻的必要性也随之出现。传统方法的一个难题是识别虚假新闻。近年来，神经网络模型由于其卓越的特征提取能力，性能已超越经典机器学习方法。本研究提出了基于深度学习的假新闻检测（DeepFND）。该技术具有视觉几何组19（VGG - 19）和双向长短期记忆（Bi - LSTM）集成模型，用于识别通过社交媒体传播的错误信息。该系统使用集成深度学习（DL）策略从文章的文本和图片中提取特征。联合特征提取器和注意力模块与一种集成方法一起使用，包括预训练和微调阶段。在本文中，我们使用了一种独特的定制损失函数。在这项研究中，我们研究了在无人干预的情况下检测互联网上虚假新闻的方法。我们使用微博、说谎者、PHEME、真假新闻和Buzzfeed数据集来分析虚假新闻和真实新闻。对多种识别虚假新闻的方法进行了比较和对比。已使用精确程序来计算所提出模型的输出。该模型99.88%的准确率优于预期。
