引用本文
  • 黄琪,陈奥,甘宏宇,罗文兵,王明文,杨兰建.基于情感感知的多重注意力网络虚假新闻检测[J].信息安全学报,已采用    [点击复制]
  • huangqi,chenao,Gan,luowenbing,wangmingwen,yanglanjian.Emotion Aware Multi-Attention Network for Fake News Detection[J].Journal of Cyber Security,Accept   [点击复制]
【打印本页】 【下载PDF全文】 查看/发表评论下载PDF阅读器关闭

过刊浏览    高级检索

本文已被:浏览 148次   下载 0  
基于情感感知的多重注意力网络虚假新闻检测
黄琪, 陈奥, 甘宏宇, 罗文兵, 王明文, 杨兰建
0
(江西师范大学)
摘要:
摘要:虚假新闻检测是应对社交媒体平台上虚假信息泛滥的关键任务,虚假新闻的传播对社会与政治稳定构成严重威胁。虽然现有的虚假新闻检测方法在捕获新闻语义和情感特征方面已取得重大进展,但大多数方法忽略了新闻信息之间的语义关联及其语义认知情感信息的相关程度。为解决上述问题,该文提出一种基于情感感知的多重注意力网络EAMAN-FND用于虚假新闻检测。受人们在判断新闻真伪时会反复阅读新闻内容及用户评论信息这一行为的启发,该文提出语义感知关联层来模拟人们反复阅读新闻内容和用户评论的过程,通过三种新闻阅读模式来揭示新闻内容与用户评论之间的深层语义关联。此外,为充分挖掘新闻信息中情感和语义间的有效信息,并充分利用它们之间的潜在关联,该文提出情感认知融合层从新闻内容和用户评论中提取并深度理解情感特征与语义特征的内在联系,通过情感分析和语义编码相结合的方式,不仅能够捕捉文本中的情感极性、强度、词汇和情绪类别等情感特征,还能识别语义层面上的关键特征并精细地融合情感和语义特征以提高虚假新闻的识别能力。该文提出的模型EAMAN-FND在公开数据集上进行了实验,相较于最先进的基准模型在准确性上分别提升了1.7%和1.2%。实验结果表明情感与语义信息融合对提升虚假新闻检测性能具有重要意义。
关键词:  虚假新闻检测  情感信息  注意力网络  特征融合
DOI:
投稿时间:2024-09-21修订日期:2025-02-18
基金项目:国家自然科学基金项目(面上项目,重点项目,重大项目)
Emotion Aware Multi-Attention Network for Fake News Detection
huangqi, chenao, Gan, luowenbing, wangmingwen, yanglanjian
(Jiangxi Normal University)
Abstract:
Abstract: Fake news detection is a key task for spreading false information on social media platforms. The spread of false news poses a serious threat to social and political stability, especially in the era of global information dissemi-nation, the spread of false information cannot be ignored. Although existing fake news detection methods have made significant progress in capturing the semantic and emotional features of news, most methods ignore the semantic associations between news information and the correlation between semantic cognitive and emotional information. To solve the above problems, this paper proposes a multiple attention network based on emotion perception EAMAN-FND for fake news detection. Inspired by the fact that people repeatedly read news content and user comments when judging the authenticity of the news, this paper proposes a semantic-aware association layer to simulate the process of people repeatedly reading news content and user comments and reveals the deep semantic associations between news content and user comments through three news reading modes. In addition, to fully explore the effective information between emotions and semantics in news information and make full use of the potential relationship between them, this paper proposes an emotional cognitive fusion layer to ex-tract and deeply understand the intrinsic relationship between emotional features and semantic features from news content and user comments. By combining sentiment analysis and semantic encoding, it can not only capture emotional features such as sentiment polarity, intensity, vocabulary, and emotional categories in the text but also identify key features at the semantic level and finely integrate these emotional and semantic features to improve the ability to identify fake news. The proposed EAMAN-FND has experimented on a public dataset, and the accuracy was improved by 1.7% and 1.2% respectively compared with the most advanced baseline model. Experimental results show that the fusion of sentiment and semantic information is of great significance in improving the performance of fake news detection.
Key words:  Fake News Detection  Emotion Information  Attention Network  Feature Fusion