Fake News Stance Detection Using Deep Learning Architecture (CNN)

Computer Engineering

In the contemporary landscape of information dissemination, the proliferation of fake news poses a significant challenge, impacting societal discourse and decision-making processes. Leveraging advancements in deep learning techniques, particularly Convolutional Neural Networks (CNNs), has emerged as a promising approach for discerning the authenticity of news content. This paper introduces a novel two-phase benchmark model, termed WELFake, designed for fake news detection. The first phase involves preprocessing the dataset and validating news content veracity through linguistic features, while the second phase integrates linguistic feature sets with word embedding (WE) and employs a voting classification scheme. To evaluate the efficacy of our approach, we meticulously curate the WELFake dataset comprising approximately 72,000 articles, amalgamating various datasets to ensure unbiased classification outcomes. Experimental results demonstrate that the WELFake model achieves a remarkable classification accuracy of 96.73%, representing a significant enhancement over existing methodologies. Specifically, our model surpasses the accuracy of bidirectional encoder representations from transformer (BERT) by 1.31% and outperforms CNN models by 4.25%. Moreover, comparative analysis with predictive-based approaches utilizing the Word2vec word embedding method showcases an improvement of up to 1.73%. The findings underscore the effectiveness of our frequency-based and focused analysis of writing patterns, affirming the utility of the WELFake model in combating the dissemination of fake news in real-time social media environments. Additionally, the model achieves an overall accuracy of 93%.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.