Scholarlink

<p>The continuous persistence of fake news harms public trust and distorts online information ecosystems. For reliable and interpretable fake news detection, this paper proposes a lightweight hybrid deep learning model that integrates Convolutional Neural Network (<b>CNN</b>), Bidirectional Long Short-Term Memory (<b>BiLSTM</b>) networks, sentiment polarity features, and a Naïve Bayes classifier. First, the CNN–BiLSTM encodes both local and sequential textual patterns, and lexicon-based sentiment cues capture emotional signals usually carried by misinformation. Second, Naïve Bayes brings a probabilistic and robust decision layer over the fused deep and sentiment representations. Extensive experiments on two benchmark datasets, Kaggle Fake News and FakeNewsNet, in a five-fold cross-validation setting, are performed. It achieves 83.17% accuracy on Kaggle and 84.04% on FakeNewsNet. The performances indicate a strong generalization capability across domains. Moreover, this architecture is far more lightweight than the transformer baselines such as Bidirectional Encoder Representations from Transformers (<b>BERT</b>) and Robustly Optimized Bidirectional Encoder Representations from Transformers (<b>RoBERTa</b>). Comprehensive ablations have been conducted to confirm the complementary contribution of each module. Therefore, it can be concluded that fusing deep semantic, contextual, and emotional signals retains great interpretability for misinformation detection and can serve as an alternative choice for resource-constrained environments. </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

An Intelligent Hybrid Deep Learning Model for Fake News Detection Based on CNN and Bi-LSTM

  • 期刊论文

  • 发表日期：

    2026年4月30日

查看全文
  • Vishakha Chauhan,
  • Abhilasha Singh,
  • Manish Bharadwaj

摘要

The continuous persistence of fake news harms public trust and distorts online information ecosystems. For reliable and interpretable fake news detection, this paper proposes a lightweight hybrid deep learning model that integrates Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM) networks, sentiment polarity features, and a Naïve Bayes classifier. First, the CNN–BiLSTM encodes both local and sequential textual patterns, and lexicon-based sentiment cues capture emotional signals usually carried by misinformation. Second, Naïve Bayes brings a probabilistic and robust decision layer over the fused deep and sentiment representations. Extensive experiments on two benchmark datasets, Kaggle Fake News and FakeNewsNet, in a five-fold cross-validation setting, are performed. It achieves 83.17% accuracy on Kaggle and 84.04% on FakeNewsNet. The performances indicate a strong generalization capability across domains. Moreover, this architecture is far more lightweight than the transformer baselines such as Bidirectional Encoder Representations from Transformers (BERT) and Robustly Optimized Bidirectional Encoder Representations from Transformers (RoBERTa). Comprehensive ablations have been conducted to confirm the complementary contribution of each module. Therefore, it can be concluded that fusing deep semantic, contextual, and emotional signals retains great interpretability for misinformation detection and can serve as an alternative choice for resource-constrained environments.