Unveiling Fake News with an Attention-Based CNN Model Using Hybrid Features

Abstract

Recent explosion of false news on social media websites has created a monument challenge to the integrity of information in the digital age and therefore sophis- ticated detection mechanisms are needed. The pragmatic problem of the dissemi- nation of misleading information calls for effective systems that can identify and counter the propagation of fake news. In this paper, we propose a new approach for the fake news detection based on hybrid embedddind techniques and deep learning architectures. We combine CountVectorizer and Global Vectors for Word Repre- sentation (GloVe) embeddings and take advantage of both statistical patterns and semantic correlations appeared in the texts. We enhance the comprehension and classification processes of complex textual content using Attention-Based Convo- lutional Neural Network (AttCNN) models with well-optimized hyper-parameters. We develop our methodology in a manner to be scalable and robust, which is ver- ified via k-fold Cross Validation (CV) and grid search hyper-parameter optimiza- tion. It has been extensively delightful and turns out to be strong and versatile across assorted false news situations. Using an AttCNN integrated with our hy- brid CV+GloVe embedding technique, we achieve state-of-the-art precision with respective final statistics of accuracy: 99.50%, F1: 99.50%, precision: 99.38% and recall: 99.62%. The proposed model is further tested using various validation set- tings to enhance its reliability for application in real-world scenarios for fake news detection.