Fake News Detection using Hybrid Deep Learning Method

Abstract

<div>As there is no single large, standard dataset available for fake news detection, this study derives a large dataset by integrating 2 publicly available datasets – Fake and real news and allData. There are 64,934 labelled news articles in the dataset. On this consolidated dataset, the Word2Vec word embedding technique yielded higher performance when combined with CNN-BiLSTM model, with accuracy, precision, recall, F1 measure and AUC-ROC values of 0.975, 0.984, 0.970, 0.977, and 0.992 respectively.</div>