The Detection of Fake News through the use of a Hierarchical Convolutional Attention Network

In the current digital era, the popularity of social media platforms has expanded tremendously. Every one of us, no matter what age or gender or class we belong to, are active on social media, sharing so much of our personal and professional life with the world. We are connected with people, countries and borders apart via our phone now. News companies are also recognizing this change in the world and are now making business through this by publishing information online. The rapid spread of an article on the internet by the normal public without checking for its authenticity is resulting in a surge in the amount of incorrect information available online. No end-user finds it simple to confirm the veracity of a given news article. These fake articles are increasing rapidly and making it far more difficult for the real information to reach the public. The purpose of this work is to develop an efficient model for assessing the credibility of a news article. This work addresses the classification problem of Fake News, given just the news article's content and the author's name. The system proposed in this paper works with Hierarchical Convolutional-Attention Network (HCAN) and Convolutional Neural Network to classify the fake news articles based on the words and phrases used in the published piece. Hierarchical Convolutional-Attention Network (HCAN) is comprised of attention-enhanced word-level and sentence-level encoders as well as a CNN to capture sequential correlation. We have also used Recurrent Neural Network (RNN) for word encoder and sentence encoder in our system. We carried out the experiment on our system using a dataset and compared the results with various baseline models like CNN, RCNN, BERT or LR. Extensive trials demonstrate that HCAN outperforms the most advanced baseline models on a collected dataset. Results of our experiment shows that our model is achieving the F1 score of 0.98.