Scholarlink

<p>Social media has a significant impact on how individuals form their opinions and make decisions. As important as it is to raise awareness, it is also crucial to make sure that the shared content is authentic. Numerous techniques have been used in the past to detect false information on social media, however they frequently have shortcomings including low effectiveness and overfitting. In order to improve the effectiveness of false news detection, we present a unique method in this paper that combines TF-IDFs and Count Vectorization with Bi-directional Gated Recurrent Units (Bi-GRU). In order to successfully handle the over fitting problem, we also incorporate a Self-Attention technique to retrieve pertinent text features from multiple contexts recorded by Bi-GRU. Using the Buzz News dataset, we assess our method by categorizing news items as authentic or fraudulent. The experimental findings, which yielded an accuracy of 0.8928, show how successful our system is. This is superior to current methods, such as Supervised Artificial Intelligence Algorithms and Deep Neural Networks (DNN), which achieved accuracies of 0.655 and 0.8649 on the Buzz News dataset, respectively.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Synergistic Approach for Fake News Detection: Bi-GRUs Coupled with Count Vectorizer and TF-IDFs

  • 期刊论文

  • 发表日期：

    2025年4月28日

查看全文
  • Thilak Bellam,
  • P. Lakshmi Prasanna

摘要

Social media has a significant impact on how individuals form their opinions and make decisions. As important as it is to raise awareness, it is also crucial to make sure that the shared content is authentic. Numerous techniques have been used in the past to detect false information on social media, however they frequently have shortcomings including low effectiveness and overfitting. In order to improve the effectiveness of false news detection, we present a unique method in this paper that combines TF-IDFs and Count Vectorization with Bi-directional Gated Recurrent Units (Bi-GRU). In order to successfully handle the over fitting problem, we also incorporate a Self-Attention technique to retrieve pertinent text features from multiple contexts recorded by Bi-GRU. Using the Buzz News dataset, we assess our method by categorizing news items as authentic or fraudulent. The experimental findings, which yielded an accuracy of 0.8928, show how successful our system is. This is superior to current methods, such as Supervised Artificial Intelligence Algorithms and Deep Neural Networks (DNN), which achieved accuracies of 0.655 and 0.8649 on the Buzz News dataset, respectively.