Fake News Detection using Deep Neural Networks

This project is part of a challenge which can be found at http://www.fakenewschallenge.org/

By Vyas Anirudh Akundy & Atish Harish Telang Patil

• Fake news is a nagging annoyance these days which can lead to the spread of misleading and fabricated information. The task of assessing the veracity of a news article is a complicated one.
• However, a first step that can be taken in identifying fake news is to detect the stance of two pieces of text, i.e estimating the relative perspective of two pieces of text

  We use Natural Language Processing and Machine Learning to tackle this problem.

• We have improved upon a previously developed baseline model whose repository can be found here

BERT(Bidirectional Encoder Representations from Transformers) language model was used to embed the text data into vector representations

• We have used bert-as-service to obtain the embeddings which can be found here

The image below, taken from the official website, clearly illustrates the task. A news headline and article are taken and the relation between them is classified into 4 classes – (unrelated, agrees, disagrees, discusses)

The neural network architecture we built is shown below