FAStan: Approaches for Fake News Detection by Stance Classification
January 2020 – June 2020
With the advent of social media, an enormous amount of information is spread over the internet. Some of this information might be fake news and rumours. Most of the people, nowadays, consume news from social media. The journalists have made several attempts to mitigate the problem of fake news, but, they face several challenges like absence of resources for validation of news and impracticality of manual validation. Hence, it becomes important to tackle the problem from an automatic perspective. I, under the supervision of Dr. Joydeep Chandra, Assistant Professor, IIT Patna, worked on proposing machine learning and deep learning based frameworks to capture both the structure of news article, identify the latent factors that play a significant role in determining fake news and handling the huge imbalance in the dataset. We propose three frameworks for the same (a) feature-based learning framework that utilizes both content and contextual features, (b) unified framework that uses pointer generator framework to segregrate between related and unrelated class and further uses features to differentiate between different classes of related class, (c) an autoencoder based framework that automatically creates an embedding of the news article.